How to free up space on disc after removing entries from IgniteCache with enabled PDS?

[daradurvs]

Igniters, we have faced with the following problem on one of our deployments.

Let's imagine that we have used IgniteCache with enabled PDS during the time:
- hardware disc space has been occupied during growing up of an amount
of data, e.g. 100Gb;
- then, we removed non-actual data, e.g 50Gb, which became useless for us;
- disc space stopped growing up with new data, but it was not
released, and still took 100Gb, instead of expected 50Gb;

Another use case:
- a user extracts data from IgniteCache to store it in separate
IgniteCache or another store;
- disc still is occupied and the user is not able to store data in the
different cache at the same cluster because of disc limitation;

How can we help the user to free up the disc space, if an amount of
data in IgniteCache has been reduced many times and will not be
increased in the nearest future?

AFAIK, we have mechanics of reusing memory pages, that allows us to
use pages which have been allocated and stored removed data for
storing new data.
Are there any chances to shrink data and free up space on disc (with
defragmentation if possible)?

--
Best Regards, Vyacheslav D.

[Vladimir Ozerov]

Hi Vyacheslav,

AFAIK this is not implemented. Shrinking/defragmentation is important
optimization. Not only because it releases free space, but also because it
decreases total number of pages. But is it not very easy to implement, as
you have to both reshuffle data entries and index entries, maintaining
consistency for concurrent reads and updates at the same time. Or
alternatively we can think of offline defragmentation. It will be easier to
implement and faster, but concurrent operations will be prohibited.

On Fri, Dec 28, 2018 at 4:08 PM Vyacheslav Daradur <[hidden email]>
wrote:

> Igniters, we have faced with the following problem on one of our
> deployments.
>
> Let's imagine that we have used IgniteCache with enabled PDS during the
> time:
> - hardware disc space has been occupied during growing up of an amount
> of data, e.g. 100Gb;
> - then, we removed non-actual data, e.g 50Gb, which became useless for us;
> - disc space stopped growing up with new data, but it was not
> released, and still took 100Gb, instead of expected 50Gb;
>
> Another use case:
> - a user extracts data from IgniteCache to store it in separate
> IgniteCache or another store;
> - disc still is occupied and the user is not able to store data in the
> different cache at the same cluster because of disc limitation;
>
> How can we help the user to free up the disc space, if an amount of
> data in IgniteCache has been reduced many times and will not be
> increased in the nearest future?
>
> AFAIK, we have mechanics of reusing memory pages, that allows us to
> use pages which have been allocated and stored removed data for
> storing new data.
> Are there any chances to shrink data and free up space on disc (with
> defragmentation if possible)?
>
> --
> Best Regards, Vyacheslav D.
>

[dmagda]

Shouldn't the OS care of defragmentation? What we need to do is to give a
way to remove stale data and "release" the allocated space somehow through
the tools, MBeans or API methods.

--
Denis


On Fri, Dec 28, 2018 at 6:24 AM Vladimir Ozerov <[hidden email]>
wrote:

> Hi Vyacheslav,
>
> AFAIK this is not implemented. Shrinking/defragmentation is important
> optimization. Not only because it releases free space, but also because it
> decreases total number of pages. But is it not very easy to implement, as
> you have to both reshuffle data entries and index entries, maintaining
> consistency for concurrent reads and updates at the same time. Or
> alternatively we can think of offline defragmentation. It will be easier to
> implement and faster, but concurrent operations will be prohibited.
>
> On Fri, Dec 28, 2018 at 4:08 PM Vyacheslav Daradur <[hidden email]>
> wrote:
>
> > Igniters, we have faced with the following problem on one of our
> > deployments.
> >
> > Let's imagine that we have used IgniteCache with enabled PDS during the
> > time:
> > - hardware disc space has been occupied during growing up of an amount
> > of data, e.g. 100Gb;
> > - then, we removed non-actual data, e.g 50Gb, which became useless for
> us;
> > - disc space stopped growing up with new data, but it was not
> > released, and still took 100Gb, instead of expected 50Gb;
> >
> > Another use case:
> > - a user extracts data from IgniteCache to store it in separate
> > IgniteCache or another store;
> > - disc still is occupied and the user is not able to store data in the
> > different cache at the same cluster because of disc limitation;
> >
> > How can we help the user to free up the disc space, if an amount of
> > data in IgniteCache has been reduced many times and will not be
> > increased in the nearest future?
> >
> > AFAIK, we have mechanics of reusing memory pages, that allows us to
> > use pages which have been allocated and stored removed data for
> > storing new data.
> > Are there any chances to shrink data and free up space on disc (with
> > defragmentation if possible)?
> >
> > --
> > Best Regards, Vyacheslav D.
> >
>

[Ivan Pavlukhin]

I suppose it is about Ignite Page Memory pages defragmentation.

We can get 100 allocated pages each of which becomes only e.g. 50%
filled after removal some entries. But they will occupy a space for
100 pages on a hard drive.

пт, 28 дек. 2018 г. в 20:45, Denis Magda <[hidden email]>:

>
> Shouldn't the OS care of defragmentation? What we need to do is to give a
> way to remove stale data and "release" the allocated space somehow through
> the tools, MBeans or API methods.
>
> --
> Denis
>
>
> On Fri, Dec 28, 2018 at 6:24 AM Vladimir Ozerov <[hidden email]>
> wrote:
>
> > Hi Vyacheslav,
> >
> > AFAIK this is not implemented. Shrinking/defragmentation is important
> > optimization. Not only because it releases free space, but also because it
> > decreases total number of pages. But is it not very easy to implement, as
> > you have to both reshuffle data entries and index entries, maintaining
> > consistency for concurrent reads and updates at the same time. Or
> > alternatively we can think of offline defragmentation. It will be easier to
> > implement and faster, but concurrent operations will be prohibited.
> >
> > On Fri, Dec 28, 2018 at 4:08 PM Vyacheslav Daradur <[hidden email]>
> > wrote:
> >
> > > Igniters, we have faced with the following problem on one of our
> > > deployments.
> > >
> > > Let's imagine that we have used IgniteCache with enabled PDS during the
> > > time:
> > > - hardware disc space has been occupied during growing up of an amount
> > > of data, e.g. 100Gb;
> > > - then, we removed non-actual data, e.g 50Gb, which became useless for
> > us;
> > > - disc space stopped growing up with new data, but it was not
> > > released, and still took 100Gb, instead of expected 50Gb;
> > >
> > > Another use case:
> > > - a user extracts data from IgniteCache to store it in separate
> > > IgniteCache or another store;
> > > - disc still is occupied and the user is not able to store data in the
> > > different cache at the same cluster because of disc limitation;
> > >
> > > How can we help the user to free up the disc space, if an amount of
> > > data in IgniteCache has been reduced many times and will not be
> > > increased in the nearest future?
> > >
> > > AFAIK, we have mechanics of reusing memory pages, that allows us to
> > > use pages which have been allocated and stored removed data for
> > > storing new data.
> > > Are there any chances to shrink data and free up space on disc (with
> > > defragmentation if possible)?
> > >
> > > --
> > > Best Regards, Vyacheslav D.
> > >
> >

--
Best regards,
Ivan Pavlukhin

[daradurvs]

Yes, it's about Page Memory defragmentation.

Pages in partitions files are stored sequentially, possible, it makes
sense to defragment pages first to avoid interpages gaps since we use
pages offset to manage them.

I filled an issue [1], I hope we will be able to find resources to
solve the issue before 2.8 release.

[1] https://issues.apache.org/jira/browse/IGNITE-10862

On Sat, Dec 29, 2018 at 10:47 AM Павлухин Иван <[hidden email]> wrote:

>
> I suppose it is about Ignite Page Memory pages defragmentation.
>
> We can get 100 allocated pages each of which becomes only e.g. 50%
> filled after removal some entries. But they will occupy a space for
> 100 pages on a hard drive.
>
> пт, 28 дек. 2018 г. в 20:45, Denis Magda <[hidden email]>:
> >
> > Shouldn't the OS care of defragmentation? What we need to do is to give a
> > way to remove stale data and "release" the allocated space somehow through
> > the tools, MBeans or API methods.
> >
> > --
> > Denis
> >
> >
> > On Fri, Dec 28, 2018 at 6:24 AM Vladimir Ozerov <[hidden email]>
> > wrote:
> >
> > > Hi Vyacheslav,
> > >
> > > AFAIK this is not implemented. Shrinking/defragmentation is important
> > > optimization. Not only because it releases free space, but also because it
> > > decreases total number of pages. But is it not very easy to implement, as
> > > you have to both reshuffle data entries and index entries, maintaining
> > > consistency for concurrent reads and updates at the same time. Or
> > > alternatively we can think of offline defragmentation. It will be easier to
> > > implement and faster, but concurrent operations will be prohibited.
> > >
> > > On Fri, Dec 28, 2018 at 4:08 PM Vyacheslav Daradur <[hidden email]>
> > > wrote:
> > >
> > > > Igniters, we have faced with the following problem on one of our
> > > > deployments.
> > > >
> > > > Let's imagine that we have used IgniteCache with enabled PDS during the
> > > > time:
> > > > - hardware disc space has been occupied during growing up of an amount
> > > > of data, e.g. 100Gb;
> > > > - then, we removed non-actual data, e.g 50Gb, which became useless for
> > > us;
> > > > - disc space stopped growing up with new data, but it was not
> > > > released, and still took 100Gb, instead of expected 50Gb;
> > > >
> > > > Another use case:
> > > > - a user extracts data from IgniteCache to store it in separate
> > > > IgniteCache or another store;
> > > > - disc still is occupied and the user is not able to store data in the
> > > > different cache at the same cluster because of disc limitation;
> > > >
> > > > How can we help the user to free up the disc space, if an amount of
> > > > data in IgniteCache has been reduced many times and will not be
> > > > increased in the nearest future?
> > > >
> > > > AFAIK, we have mechanics of reusing memory pages, that allows us to
> > > > use pages which have been allocated and stored removed data for
> > > > storing new data.
> > > > Are there any chances to shrink data and free up space on disc (with
> > > > defragmentation if possible)?
> > > >
> > > > --
> > > > Best Regards, Vyacheslav D.
> > > >
> > >
>
>
>
> --
> Best regards,
> Ivan Pavlukhin

--
Best Regards, Vyacheslav D.

[Ivan Pavlukhin]

Vyacheslav,

Have you investigated how other vendors (Oracle, Postgres) tackle this problem?

I have one wild idea. Could the problem be solved by stopping a node
which need to be defragmented, clearing persistence files and
restarting the node? After rebalance the node will receive all data
back without fragmentation. I see a big downside -- sending data
across the network. But perhaps we can play with affinity and start
new node on the same host which will receive the same data, after that
old node can be stopped. It looks more as kind of workaround but
perhaps it can be turned into workable solution.

ср, 9 янв. 2019 г. в 10:49, Vyacheslav Daradur <[hidden email]>:

>
> Yes, it's about Page Memory defragmentation.
>
> Pages in partitions files are stored sequentially, possible, it makes
> sense to defragment pages first to avoid interpages gaps since we use
> pages offset to manage them.
>
> I filled an issue [1], I hope we will be able to find resources to
> solve the issue before 2.8 release.
>
> [1] https://issues.apache.org/jira/browse/IGNITE-10862
>
> On Sat, Dec 29, 2018 at 10:47 AM Павлухин Иван <[hidden email]> wrote:
> >
> > I suppose it is about Ignite Page Memory pages defragmentation.
> >
> > We can get 100 allocated pages each of which becomes only e.g. 50%
> > filled after removal some entries. But they will occupy a space for
> > 100 pages on a hard drive.
> >
> > пт, 28 дек. 2018 г. в 20:45, Denis Magda <[hidden email]>:
> > >
> > > Shouldn't the OS care of defragmentation? What we need to do is to give a
> > > way to remove stale data and "release" the allocated space somehow through
> > > the tools, MBeans or API methods.
> > >
> > > --
> > > Denis
> > >
> > >
> > > On Fri, Dec 28, 2018 at 6:24 AM Vladimir Ozerov <[hidden email]>
> > > wrote:
> > >
> > > > Hi Vyacheslav,
> > > >
> > > > AFAIK this is not implemented. Shrinking/defragmentation is important
> > > > optimization. Not only because it releases free space, but also because it
> > > > decreases total number of pages. But is it not very easy to implement, as
> > > > you have to both reshuffle data entries and index entries, maintaining
> > > > consistency for concurrent reads and updates at the same time. Or
> > > > alternatively we can think of offline defragmentation. It will be easier to
> > > > implement and faster, but concurrent operations will be prohibited.
> > > >
> > > > On Fri, Dec 28, 2018 at 4:08 PM Vyacheslav Daradur <[hidden email]>
> > > > wrote:
> > > >
> > > > > Igniters, we have faced with the following problem on one of our
> > > > > deployments.
> > > > >
> > > > > Let's imagine that we have used IgniteCache with enabled PDS during the
> > > > > time:
> > > > > - hardware disc space has been occupied during growing up of an amount
> > > > > of data, e.g. 100Gb;
> > > > > - then, we removed non-actual data, e.g 50Gb, which became useless for
> > > > us;
> > > > > - disc space stopped growing up with new data, but it was not
> > > > > released, and still took 100Gb, instead of expected 50Gb;
> > > > >
> > > > > Another use case:
> > > > > - a user extracts data from IgniteCache to store it in separate
> > > > > IgniteCache or another store;
> > > > > - disc still is occupied and the user is not able to store data in the
> > > > > different cache at the same cluster because of disc limitation;
> > > > >
> > > > > How can we help the user to free up the disc space, if an amount of
> > > > > data in IgniteCache has been reduced many times and will not be
> > > > > increased in the nearest future?
> > > > >
> > > > > AFAIK, we have mechanics of reusing memory pages, that allows us to
> > > > > use pages which have been allocated and stored removed data for
> > > > > storing new data.
> > > > > Are there any chances to shrink data and free up space on disc (with
> > > > > defragmentation if possible)?
> > > > >
> > > > > --
> > > > > Best Regards, Vyacheslav D.
> > > > >
> > > >
> >
> >
> >
> > --
> > Best regards,
> > Ivan Pavlukhin
>
>
>
> --
> Best Regards, Vyacheslav D.

--
Best regards,
Ivan Pavlukhin

[Dmitry Pavlov]

In the TC Bot, I used to create the second cache with CacheV2 name and
migrate needed data from Cache  V1 to V2.

After CacheV1 destroy(), files are removed and disk space is freed.

ср, 9 янв. 2019 г. в 12:04, Павлухин Иван <[hidden email]>:

> Vyacheslav,
>
> Have you investigated how other vendors (Oracle, Postgres) tackle this
> problem?
>
> I have one wild idea. Could the problem be solved by stopping a node
> which need to be defragmented, clearing persistence files and
> restarting the node? After rebalance the node will receive all data
> back without fragmentation. I see a big downside -- sending data
> across the network. But perhaps we can play with affinity and start
> new node on the same host which will receive the same data, after that
> old node can be stopped. It looks more as kind of workaround but
> perhaps it can be turned into workable solution.
>
> ср, 9 янв. 2019 г. в 10:49, Vyacheslav Daradur <[hidden email]>:
> >
> > Yes, it's about Page Memory defragmentation.
> >
> > Pages in partitions files are stored sequentially, possible, it makes
> > sense to defragment pages first to avoid interpages gaps since we use
> > pages offset to manage them.
> >
> > I filled an issue [1], I hope we will be able to find resources to
> > solve the issue before 2.8 release.
> >
> > [1] https://issues.apache.org/jira/browse/IGNITE-10862
> >
> > On Sat, Dec 29, 2018 at 10:47 AM Павлухин Иван <[hidden email]>
> wrote:
> > >
> > > I suppose it is about Ignite Page Memory pages defragmentation.
> > >
> > > We can get 100 allocated pages each of which becomes only e.g. 50%
> > > filled after removal some entries. But they will occupy a space for
> > > 100 pages on a hard drive.
> > >
> > > пт, 28 дек. 2018 г. в 20:45, Denis Magda <[hidden email]>:
> > > >
> > > > Shouldn't the OS care of defragmentation? What we need to do is to
> give a
> > > > way to remove stale data and "release" the allocated space somehow
> through
> > > > the tools, MBeans or API methods.
> > > >
> > > > --
> > > > Denis
> > > >
> > > >
> > > > On Fri, Dec 28, 2018 at 6:24 AM Vladimir Ozerov <
> [hidden email]>
> > > > wrote:
> > > >
> > > > > Hi Vyacheslav,
> > > > >
> > > > > AFAIK this is not implemented. Shrinking/defragmentation is
> important
> > > > > optimization. Not only because it releases free space, but also
> because it
> > > > > decreases total number of pages. But is it not very easy to
> implement, as
> > > > > you have to both reshuffle data entries and index entries,
> maintaining
> > > > > consistency for concurrent reads and updates at the same time. Or
> > > > > alternatively we can think of offline defragmentation. It will be
> easier to
> > > > > implement and faster, but concurrent operations will be prohibited.
> > > > >
> > > > > On Fri, Dec 28, 2018 at 4:08 PM Vyacheslav Daradur <
> [hidden email]>
> > > > > wrote:
> > > > >
> > > > > > Igniters, we have faced with the following problem on one of our
> > > > > > deployments.
> > > > > >
> > > > > > Let's imagine that we have used IgniteCache with enabled PDS
> during the
> > > > > > time:
> > > > > > - hardware disc space has been occupied during growing up of an
> amount
> > > > > > of data, e.g. 100Gb;
> > > > > > - then, we removed non-actual data, e.g 50Gb, which became
> useless for
> > > > > us;
> > > > > > - disc space stopped growing up with new data, but it was not
> > > > > > released, and still took 100Gb, instead of expected 50Gb;
> > > > > >
> > > > > > Another use case:
> > > > > > - a user extracts data from IgniteCache to store it in separate
> > > > > > IgniteCache or another store;
> > > > > > - disc still is occupied and the user is not able to store data
> in the
> > > > > > different cache at the same cluster because of disc limitation;
> > > > > >
> > > > > > How can we help the user to free up the disc space, if an amount
> of
> > > > > > data in IgniteCache has been reduced many times and will not be
> > > > > > increased in the nearest future?
> > > > > >
> > > > > > AFAIK, we have mechanics of reusing memory pages, that allows us
> to
> > > > > > use pages which have been allocated and stored removed data for
> > > > > > storing new data.
> > > > > > Are there any chances to shrink data and free up space on disc
> (with
> > > > > > defragmentation if possible)?
> > > > > >
> > > > > > --
> > > > > > Best Regards, Vyacheslav D.
> > > > > >
> > > > >
> > >
> > >
> > >
> > > --
> > > Best regards,
> > > Ivan Pavlukhin
> >
> >
> >
> > --
> > Best Regards, Vyacheslav D.
>
>
>
> --
> Best regards,
> Ivan Pavlukhin
>

[Anton Vinogradov-2]

Dmitriy,

This does not look like a production-ready case :)

How about
1) Once you need to write an entry - you have to chose not random "page
from free-list with enough space"
but "page from free-list with enough space closest to the beginning of the
file".

2) Once you remove entry you have to merge the rest of the entries at this
page to the
"page from free-list with enough space closest to the beginning of the file"
if possible. (optional)

3) Partition file tail with empty pages can bу removed at any time.

4) In case you have cold data inside the tail, just lock the page and
perform migration to
"page from free-list with enough space closest to the beginning of the
file".
This operation can be scheduled.

On Wed, Jan 9, 2019 at 4:43 PM Dmitriy Pavlov <[hidden email]> wrote:

> In the TC Bot, I used to create the second cache with CacheV2 name and
> migrate needed data from Cache  V1 to V2.
>
> After CacheV1 destroy(), files are removed and disk space is freed.
>
> ср, 9 янв. 2019 г. в 12:04, Павлухин Иван <[hidden email]>:
>
> > Vyacheslav,
> >
> > Have you investigated how other vendors (Oracle, Postgres) tackle this
> > problem?
> >
> > I have one wild idea. Could the problem be solved by stopping a node
> > which need to be defragmented, clearing persistence files and
> > restarting the node? After rebalance the node will receive all data
> > back without fragmentation. I see a big downside -- sending data
> > across the network. But perhaps we can play with affinity and start
> > new node on the same host which will receive the same data, after that
> > old node can be stopped. It looks more as kind of workaround but
> > perhaps it can be turned into workable solution.
> >
> > ср, 9 янв. 2019 г. в 10:49, Vyacheslav Daradur <[hidden email]>:
> > >
> > > Yes, it's about Page Memory defragmentation.
> > >
> > > Pages in partitions files are stored sequentially, possible, it makes
> > > sense to defragment pages first to avoid interpages gaps since we use
> > > pages offset to manage them.
> > >
> > > I filled an issue [1], I hope we will be able to find resources to
> > > solve the issue before 2.8 release.
> > >
> > > [1] https://issues.apache.org/jira/browse/IGNITE-10862
> > >
> > > On Sat, Dec 29, 2018 at 10:47 AM Павлухин Иван <[hidden email]>
> > wrote:
> > > >
> > > > I suppose it is about Ignite Page Memory pages defragmentation.
> > > >
> > > > We can get 100 allocated pages each of which becomes only e.g. 50%
> > > > filled after removal some entries. But they will occupy a space for
> > > > 100 pages on a hard drive.
> > > >
> > > > пт, 28 дек. 2018 г. в 20:45, Denis Magda <[hidden email]>:
> > > > >
> > > > > Shouldn't the OS care of defragmentation? What we need to do is to
> > give a
> > > > > way to remove stale data and "release" the allocated space somehow
> > through
> > > > > the tools, MBeans or API methods.
> > > > >
> > > > > --
> > > > > Denis
> > > > >
> > > > >
> > > > > On Fri, Dec 28, 2018 at 6:24 AM Vladimir Ozerov <
> > [hidden email]>
> > > > > wrote:
> > > > >
> > > > > > Hi Vyacheslav,
> > > > > >
> > > > > > AFAIK this is not implemented. Shrinking/defragmentation is
> > important
> > > > > > optimization. Not only because it releases free space, but also
> > because it
> > > > > > decreases total number of pages. But is it not very easy to
> > implement, as
> > > > > > you have to both reshuffle data entries and index entries,
> > maintaining
> > > > > > consistency for concurrent reads and updates at the same time. Or
> > > > > > alternatively we can think of offline defragmentation. It will be
> > easier to
> > > > > > implement and faster, but concurrent operations will be
> prohibited.
> > > > > >
> > > > > > On Fri, Dec 28, 2018 at 4:08 PM Vyacheslav Daradur <
> > [hidden email]>
> > > > > > wrote:
> > > > > >
> > > > > > > Igniters, we have faced with the following problem on one of
> our
> > > > > > > deployments.
> > > > > > >
> > > > > > > Let's imagine that we have used IgniteCache with enabled PDS
> > during the
> > > > > > > time:
> > > > > > > - hardware disc space has been occupied during growing up of an
> > amount
> > > > > > > of data, e.g. 100Gb;
> > > > > > > - then, we removed non-actual data, e.g 50Gb, which became
> > useless for
> > > > > > us;
> > > > > > > - disc space stopped growing up with new data, but it was not
> > > > > > > released, and still took 100Gb, instead of expected 50Gb;
> > > > > > >
> > > > > > > Another use case:
> > > > > > > - a user extracts data from IgniteCache to store it in separate
> > > > > > > IgniteCache or another store;
> > > > > > > - disc still is occupied and the user is not able to store data
> > in the
> > > > > > > different cache at the same cluster because of disc limitation;
> > > > > > >
> > > > > > > How can we help the user to free up the disc space, if an
> amount
> > of
> > > > > > > data in IgniteCache has been reduced many times and will not be
> > > > > > > increased in the nearest future?
> > > > > > >
> > > > > > > AFAIK, we have mechanics of reusing memory pages, that allows
> us
> > to
> > > > > > > use pages which have been allocated and stored removed data for
> > > > > > > storing new data.
> > > > > > > Are there any chances to shrink data and free up space on disc
> > (with
> > > > > > > defragmentation if possible)?
> > > > > > >
> > > > > > > --
> > > > > > > Best Regards, Vyacheslav D.
> > > > > > >
> > > > > >
> > > >
> > > >
> > > >
> > > > --
> > > > Best regards,
> > > > Ivan Pavlukhin
> > >
> > >
> > >
> > > --
> > > Best Regards, Vyacheslav D.
> >
> >
> >
> > --
> > Best regards,
> > Ivan Pavlukhin
> >
>

[dmagda]

The issue starts hitting others who deploy Ignite persistence in production:
https://issues.apache.org/jira/browse/IGNITE-12152

Alex, I'm curious is this a fundamental problem. Asked the same question in
JIRA but, probably, this discussion is a better place to get to the bottom
first:
https://issues.apache.org/jira/browse/IGNITE-10862

-
Denis


On Thu, Jan 10, 2019 at 6:01 AM Anton Vinogradov <[hidden email]> wrote:

> Dmitriy,
>
> This does not look like a production-ready case :)
>
> How about
> 1) Once you need to write an entry - you have to chose not random "page
> from free-list with enough space"
> but "page from free-list with enough space closest to the beginning of the
> file".
>
> 2) Once you remove entry you have to merge the rest of the entries at this
> page to the
> "page from free-list with enough space closest to the beginning of the
> file"
> if possible. (optional)
>
> 3) Partition file tail with empty pages can bу removed at any time.
>
> 4) In case you have cold data inside the tail, just lock the page and
> perform migration to
> "page from free-list with enough space closest to the beginning of the
> file".
> This operation can be scheduled.
>
> On Wed, Jan 9, 2019 at 4:43 PM Dmitriy Pavlov <[hidden email]> wrote:
>
> > In the TC Bot, I used to create the second cache with CacheV2 name and
> > migrate needed data from Cache  V1 to V2.
> >
> > After CacheV1 destroy(), files are removed and disk space is freed.
> >
> > ср, 9 янв. 2019 г. в 12:04, Павлухин Иван <[hidden email]>:
> >
> > > Vyacheslav,
> > >
> > > Have you investigated how other vendors (Oracle, Postgres) tackle this
> > > problem?
> > >
> > > I have one wild idea. Could the problem be solved by stopping a node
> > > which need to be defragmented, clearing persistence files and
> > > restarting the node? After rebalance the node will receive all data
> > > back without fragmentation. I see a big downside -- sending data
> > > across the network. But perhaps we can play with affinity and start
> > > new node on the same host which will receive the same data, after that
> > > old node can be stopped. It looks more as kind of workaround but
> > > perhaps it can be turned into workable solution.
> > >
> > > ср, 9 янв. 2019 г. в 10:49, Vyacheslav Daradur <[hidden email]>:
> > > >
> > > > Yes, it's about Page Memory defragmentation.
> > > >
> > > > Pages in partitions files are stored sequentially, possible, it makes
> > > > sense to defragment pages first to avoid interpages gaps since we use
> > > > pages offset to manage them.
> > > >
> > > > I filled an issue [1], I hope we will be able to find resources to
> > > > solve the issue before 2.8 release.
> > > >
> > > > [1] https://issues.apache.org/jira/browse/IGNITE-10862
> > > >
> > > > On Sat, Dec 29, 2018 at 10:47 AM Павлухин Иван <[hidden email]>
> > > wrote:
> > > > >
> > > > > I suppose it is about Ignite Page Memory pages defragmentation.
> > > > >
> > > > > We can get 100 allocated pages each of which becomes only e.g. 50%
> > > > > filled after removal some entries. But they will occupy a space for
> > > > > 100 pages on a hard drive.
> > > > >
> > > > > пт, 28 дек. 2018 г. в 20:45, Denis Magda <[hidden email]>:
> > > > > >
> > > > > > Shouldn't the OS care of defragmentation? What we need to do is
> to
> > > give a
> > > > > > way to remove stale data and "release" the allocated space
> somehow
> > > through
> > > > > > the tools, MBeans or API methods.
> > > > > >
> > > > > > --
> > > > > > Denis
> > > > > >
> > > > > >
> > > > > > On Fri, Dec 28, 2018 at 6:24 AM Vladimir Ozerov <
> > > [hidden email]>
> > > > > > wrote:
> > > > > >
> > > > > > > Hi Vyacheslav,
> > > > > > >
> > > > > > > AFAIK this is not implemented. Shrinking/defragmentation is
> > > important
> > > > > > > optimization. Not only because it releases free space, but also
> > > because it
> > > > > > > decreases total number of pages. But is it not very easy to
> > > implement, as
> > > > > > > you have to both reshuffle data entries and index entries,
> > > maintaining
> > > > > > > consistency for concurrent reads and updates at the same time.
> Or
> > > > > > > alternatively we can think of offline defragmentation. It will
> be
> > > easier to
> > > > > > > implement and faster, but concurrent operations will be
> > prohibited.
> > > > > > >
> > > > > > > On Fri, Dec 28, 2018 at 4:08 PM Vyacheslav Daradur <
> > > [hidden email]>
> > > > > > > wrote:
> > > > > > >
> > > > > > > > Igniters, we have faced with the following problem on one of
> > our
> > > > > > > > deployments.
> > > > > > > >
> > > > > > > > Let's imagine that we have used IgniteCache with enabled PDS
> > > during the
> > > > > > > > time:
> > > > > > > > - hardware disc space has been occupied during growing up of
> an
> > > amount
> > > > > > > > of data, e.g. 100Gb;
> > > > > > > > - then, we removed non-actual data, e.g 50Gb, which became
> > > useless for
> > > > > > > us;
> > > > > > > > - disc space stopped growing up with new data, but it was not
> > > > > > > > released, and still took 100Gb, instead of expected 50Gb;
> > > > > > > >
> > > > > > > > Another use case:
> > > > > > > > - a user extracts data from IgniteCache to store it in
> separate
> > > > > > > > IgniteCache or another store;
> > > > > > > > - disc still is occupied and the user is not able to store
> data
> > > in the
> > > > > > > > different cache at the same cluster because of disc
> limitation;
> > > > > > > >
> > > > > > > > How can we help the user to free up the disc space, if an
> > amount
> > > of
> > > > > > > > data in IgniteCache has been reduced many times and will not
> be
> > > > > > > > increased in the nearest future?
> > > > > > > >
> > > > > > > > AFAIK, we have mechanics of reusing memory pages, that allows
> > us
> > > to
> > > > > > > > use pages which have been allocated and stored removed data
> for
> > > > > > > > storing new data.
> > > > > > > > Are there any chances to shrink data and free up space on
> disc
> > > (with
> > > > > > > > defragmentation if possible)?
> > > > > > > >
> > > > > > > > --
> > > > > > > > Best Regards, Vyacheslav D.
> > > > > > > >
> > > > > > >
> > > > >
> > > > >
> > > > >
> > > > > --
> > > > > Best regards,
> > > > > Ivan Pavlukhin
> > > >
> > > >
> > > >
> > > > --
> > > > Best Regards, Vyacheslav D.
> > >
> > >
> > >
> > > --
> > > Best regards,
> > > Ivan Pavlukhin
> > >
> >
>

[Alexey Goncharuk]

Denis,

It's not fundamental, but quite complex. In postgres, for example, this is
not maintained automatically and store compaction is performed using the
full vacuum command, which acquires exclusive table lock, so no concurrent
activities on the table are possible.

The solution which Anton suggested does not look easy because it will most
likely significantly hurt performance: it is hard to maintain a data
structure to choose "page from free-list with enough space closest to the
beginning of the file". Overall, we can think of something similar to
postgres, when a space can be freed in some maintenance mode.

Online space cleanup sounds tricky for me, or at least I cannot think about
a plausible solution right away.

--AG

пт, 13 сент. 2019 г. в 19:43, Denis Magda <[hidden email]>:

> The issue starts hitting others who deploy Ignite persistence in
> production:
> https://issues.apache.org/jira/browse/IGNITE-12152
>
> Alex, I'm curious is this a fundamental problem. Asked the same question in
> JIRA but, probably, this discussion is a better place to get to the bottom
> first:
> https://issues.apache.org/jira/browse/IGNITE-10862
>
> -
> Denis
>
>
> On Thu, Jan 10, 2019 at 6:01 AM Anton Vinogradov <[hidden email]> wrote:
>
> > Dmitriy,
> >
> > This does not look like a production-ready case :)
> >
> > How about
> > 1) Once you need to write an entry - you have to chose not random "page
> > from free-list with enough space"
> > but "page from free-list with enough space closest to the beginning of
> the
> > file".
> >
> > 2) Once you remove entry you have to merge the rest of the entries at
> this
> > page to the
> > "page from free-list with enough space closest to the beginning of the
> > file"
> > if possible. (optional)
> >
> > 3) Partition file tail with empty pages can bу removed at any time.
> >
> > 4) In case you have cold data inside the tail, just lock the page and
> > perform migration to
> > "page from free-list with enough space closest to the beginning of the
> > file".
> > This operation can be scheduled.
> >
> > On Wed, Jan 9, 2019 at 4:43 PM Dmitriy Pavlov <[hidden email]>
> wrote:
> >
> > > In the TC Bot, I used to create the second cache with CacheV2 name and
> > > migrate needed data from Cache  V1 to V2.
> > >
> > > After CacheV1 destroy(), files are removed and disk space is freed.
> > >
> > > ср, 9 янв. 2019 г. в 12:04, Павлухин Иван <[hidden email]>:
> > >
> > > > Vyacheslav,
> > > >
> > > > Have you investigated how other vendors (Oracle, Postgres) tackle
> this
> > > > problem?
> > > >
> > > > I have one wild idea. Could the problem be solved by stopping a node
> > > > which need to be defragmented, clearing persistence files and
> > > > restarting the node? After rebalance the node will receive all data
> > > > back without fragmentation. I see a big downside -- sending data
> > > > across the network. But perhaps we can play with affinity and start
> > > > new node on the same host which will receive the same data, after
> that
> > > > old node can be stopped. It looks more as kind of workaround but
> > > > perhaps it can be turned into workable solution.
> > > >
> > > > ср, 9 янв. 2019 г. в 10:49, Vyacheslav Daradur <[hidden email]
> >:
> > > > >
> > > > > Yes, it's about Page Memory defragmentation.
> > > > >
> > > > > Pages in partitions files are stored sequentially, possible, it
> makes
> > > > > sense to defragment pages first to avoid interpages gaps since we
> use
> > > > > pages offset to manage them.
> > > > >
> > > > > I filled an issue [1], I hope we will be able to find resources to
> > > > > solve the issue before 2.8 release.
> > > > >
> > > > > [1] https://issues.apache.org/jira/browse/IGNITE-10862
> > > > >
> > > > > On Sat, Dec 29, 2018 at 10:47 AM Павлухин Иван <
> [hidden email]>
> > > > wrote:
> > > > > >
> > > > > > I suppose it is about Ignite Page Memory pages defragmentation.
> > > > > >
> > > > > > We can get 100 allocated pages each of which becomes only e.g.
> 50%
> > > > > > filled after removal some entries. But they will occupy a space
> for
> > > > > > 100 pages on a hard drive.
> > > > > >
> > > > > > пт, 28 дек. 2018 г. в 20:45, Denis Magda <[hidden email]>:
> > > > > > >
> > > > > > > Shouldn't the OS care of defragmentation? What we need to do is
> > to
> > > > give a
> > > > > > > way to remove stale data and "release" the allocated space
> > somehow
> > > > through
> > > > > > > the tools, MBeans or API methods.
> > > > > > >
> > > > > > > --
> > > > > > > Denis
> > > > > > >
> > > > > > >
> > > > > > > On Fri, Dec 28, 2018 at 6:24 AM Vladimir Ozerov <
> > > > [hidden email]>
> > > > > > > wrote:
> > > > > > >
> > > > > > > > Hi Vyacheslav,
> > > > > > > >
> > > > > > > > AFAIK this is not implemented. Shrinking/defragmentation is
> > > > important
> > > > > > > > optimization. Not only because it releases free space, but
> also
> > > > because it
> > > > > > > > decreases total number of pages. But is it not very easy to
> > > > implement, as
> > > > > > > > you have to both reshuffle data entries and index entries,
> > > > maintaining
> > > > > > > > consistency for concurrent reads and updates at the same
> time.
> > Or
> > > > > > > > alternatively we can think of offline defragmentation. It
> will
> > be
> > > > easier to
> > > > > > > > implement and faster, but concurrent operations will be
> > > prohibited.
> > > > > > > >
> > > > > > > > On Fri, Dec 28, 2018 at 4:08 PM Vyacheslav Daradur <
> > > > [hidden email]>
> > > > > > > > wrote:
> > > > > > > >
> > > > > > > > > Igniters, we have faced with the following problem on one
> of
> > > our
> > > > > > > > > deployments.
> > > > > > > > >
> > > > > > > > > Let's imagine that we have used IgniteCache with enabled
> PDS
> > > > during the
> > > > > > > > > time:
> > > > > > > > > - hardware disc space has been occupied during growing up
> of
> > an
> > > > amount
> > > > > > > > > of data, e.g. 100Gb;
> > > > > > > > > - then, we removed non-actual data, e.g 50Gb, which became
> > > > useless for
> > > > > > > > us;
> > > > > > > > > - disc space stopped growing up with new data, but it was
> not
> > > > > > > > > released, and still took 100Gb, instead of expected 50Gb;
> > > > > > > > >
> > > > > > > > > Another use case:
> > > > > > > > > - a user extracts data from IgniteCache to store it in
> > separate
> > > > > > > > > IgniteCache or another store;
> > > > > > > > > - disc still is occupied and the user is not able to store
> > data
> > > > in the
> > > > > > > > > different cache at the same cluster because of disc
> > limitation;
> > > > > > > > >
> > > > > > > > > How can we help the user to free up the disc space, if an
> > > amount
> > > > of
> > > > > > > > > data in IgniteCache has been reduced many times and will
> not
> > be
> > > > > > > > > increased in the nearest future?
> > > > > > > > >
> > > > > > > > > AFAIK, we have mechanics of reusing memory pages, that
> allows
> > > us
> > > > to
> > > > > > > > > use pages which have been allocated and stored removed data
> > for
> > > > > > > > > storing new data.
> > > > > > > > > Are there any chances to shrink data and free up space on
> > disc
> > > > (with
> > > > > > > > > defragmentation if possible)?
> > > > > > > > >
> > > > > > > > > --
> > > > > > > > > Best Regards, Vyacheslav D.
> > > > > > > > >
> > > > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > --
> > > > > > Best regards,
> > > > > > Ivan Pavlukhin
> > > > >
> > > > >
> > > > >
> > > > > --
> > > > > Best Regards, Vyacheslav D.
> > > >
> > > >
> > > >
> > > > --
> > > > Best regards,
> > > > Ivan Pavlukhin
> > > >
> > >
> >
>

[Anton Vinogradov-2]

Alexey,

>>  The solution which Anton suggested does not look easy because it will
most likely significantly hurt performance
Mostly agree here, but what drop do we expect? What price do we ready to
pay?
Not sure, but seems some vendors ready to pay, for example, 5% drop for
this.

>> it is hard to maintain a data structure to choose "page from free-list
with enough space closest to the beginning of the file".
We can just split each free-list bucket to the couple and use first for
pages in the first half of the file and the second for the last.
Only two buckets required here since, during the file shrink, first
bucket's window will be shrank too.
Seems, this give us the same price on put, just use the first bucket in
case it's not empty.
Remove price (with merge) will be increased, of course.

The compromise solution is to have priority put (to the first path of the
file), with keeping removal as is, and schedulable per-page migration for
the rest of the data during the low activity period.

On Wed, Sep 18, 2019 at 8:03 PM Alexey Goncharuk <[hidden email]>
wrote:

> Denis,
>
> It's not fundamental, but quite complex. In postgres, for example, this is
> not maintained automatically and store compaction is performed using the
> full vacuum command, which acquires exclusive table lock, so no concurrent
> activities on the table are possible.
>
> The solution which Anton suggested does not look easy because it will most
> likely significantly hurt performance: it is hard to maintain a data
> structure to choose "page from free-list with enough space closest to the
> beginning of the file". Overall, we can think of something similar to
> postgres, when a space can be freed in some maintenance mode.
>
> Online space cleanup sounds tricky for me, or at least I cannot think about
> a plausible solution right away.
>
> --AG
>
> пт, 13 сент. 2019 г. в 19:43, Denis Magda <[hidden email]>:
>
> > The issue starts hitting others who deploy Ignite persistence in
> > production:
> > https://issues.apache.org/jira/browse/IGNITE-12152
> >
> > Alex, I'm curious is this a fundamental problem. Asked the same question
> in
> > JIRA but, probably, this discussion is a better place to get to the
> bottom
> > first:
> > https://issues.apache.org/jira/browse/IGNITE-10862
> >
> > -
> > Denis
> >
> >
> > On Thu, Jan 10, 2019 at 6:01 AM Anton Vinogradov <[hidden email]> wrote:
> >
> > > Dmitriy,
> > >
> > > This does not look like a production-ready case :)
> > >
> > > How about
> > > 1) Once you need to write an entry - you have to chose not random "page
> > > from free-list with enough space"
> > > but "page from free-list with enough space closest to the beginning of
> > the
> > > file".
> > >
> > > 2) Once you remove entry you have to merge the rest of the entries at
> > this
> > > page to the
> > > "page from free-list with enough space closest to the beginning of the
> > > file"
> > > if possible. (optional)
> > >
> > > 3) Partition file tail with empty pages can bу removed at any time.
> > >
> > > 4) In case you have cold data inside the tail, just lock the page and
> > > perform migration to
> > > "page from free-list with enough space closest to the beginning of the
> > > file".
> > > This operation can be scheduled.
> > >
> > > On Wed, Jan 9, 2019 at 4:43 PM Dmitriy Pavlov <[hidden email]>
> > wrote:
> > >
> > > > In the TC Bot, I used to create the second cache with CacheV2 name
> and
> > > > migrate needed data from Cache  V1 to V2.
> > > >
> > > > After CacheV1 destroy(), files are removed and disk space is freed.
> > > >
> > > > ср, 9 янв. 2019 г. в 12:04, Павлухин Иван <[hidden email]>:
> > > >
> > > > > Vyacheslav,
> > > > >
> > > > > Have you investigated how other vendors (Oracle, Postgres) tackle
> > this
> > > > > problem?
> > > > >
> > > > > I have one wild idea. Could the problem be solved by stopping a
> node
> > > > > which need to be defragmented, clearing persistence files and
> > > > > restarting the node? After rebalance the node will receive all data
> > > > > back without fragmentation. I see a big downside -- sending data
> > > > > across the network. But perhaps we can play with affinity and start
> > > > > new node on the same host which will receive the same data, after
> > that
> > > > > old node can be stopped. It looks more as kind of workaround but
> > > > > perhaps it can be turned into workable solution.
> > > > >
> > > > > ср, 9 янв. 2019 г. в 10:49, Vyacheslav Daradur <
> [hidden email]
> > >:
> > > > > >
> > > > > > Yes, it's about Page Memory defragmentation.
> > > > > >
> > > > > > Pages in partitions files are stored sequentially, possible, it
> > makes
> > > > > > sense to defragment pages first to avoid interpages gaps since we
> > use
> > > > > > pages offset to manage them.
> > > > > >
> > > > > > I filled an issue [1], I hope we will be able to find resources
> to
> > > > > > solve the issue before 2.8 release.
> > > > > >
> > > > > > [1] https://issues.apache.org/jira/browse/IGNITE-10862
> > > > > >
> > > > > > On Sat, Dec 29, 2018 at 10:47 AM Павлухин Иван <
> > [hidden email]>
> > > > > wrote:
> > > > > > >
> > > > > > > I suppose it is about Ignite Page Memory pages defragmentation.
> > > > > > >
> > > > > > > We can get 100 allocated pages each of which becomes only e.g.
> > 50%
> > > > > > > filled after removal some entries. But they will occupy a space
> > for
> > > > > > > 100 pages on a hard drive.
> > > > > > >
> > > > > > > пт, 28 дек. 2018 г. в 20:45, Denis Magda <[hidden email]>:
> > > > > > > >
> > > > > > > > Shouldn't the OS care of defragmentation? What we need to do
> is
> > > to
> > > > > give a
> > > > > > > > way to remove stale data and "release" the allocated space
> > > somehow
> > > > > through
> > > > > > > > the tools, MBeans or API methods.
> > > > > > > >
> > > > > > > > --
> > > > > > > > Denis
> > > > > > > >
> > > > > > > >
> > > > > > > > On Fri, Dec 28, 2018 at 6:24 AM Vladimir Ozerov <
> > > > > [hidden email]>
> > > > > > > > wrote:
> > > > > > > >
> > > > > > > > > Hi Vyacheslav,
> > > > > > > > >
> > > > > > > > > AFAIK this is not implemented. Shrinking/defragmentation is
> > > > > important
> > > > > > > > > optimization. Not only because it releases free space, but
> > also
> > > > > because it
> > > > > > > > > decreases total number of pages. But is it not very easy to
> > > > > implement, as
> > > > > > > > > you have to both reshuffle data entries and index entries,
> > > > > maintaining
> > > > > > > > > consistency for concurrent reads and updates at the same
> > time.
> > > Or
> > > > > > > > > alternatively we can think of offline defragmentation. It
> > will
> > > be
> > > > > easier to
> > > > > > > > > implement and faster, but concurrent operations will be
> > > > prohibited.
> > > > > > > > >
> > > > > > > > > On Fri, Dec 28, 2018 at 4:08 PM Vyacheslav Daradur <
> > > > > [hidden email]>
> > > > > > > > > wrote:
> > > > > > > > >
> > > > > > > > > > Igniters, we have faced with the following problem on one
> > of
> > > > our
> > > > > > > > > > deployments.
> > > > > > > > > >
> > > > > > > > > > Let's imagine that we have used IgniteCache with enabled
> > PDS
> > > > > during the
> > > > > > > > > > time:
> > > > > > > > > > - hardware disc space has been occupied during growing up
> > of
> > > an
> > > > > amount
> > > > > > > > > > of data, e.g. 100Gb;
> > > > > > > > > > - then, we removed non-actual data, e.g 50Gb, which
> became
> > > > > useless for
> > > > > > > > > us;
> > > > > > > > > > - disc space stopped growing up with new data, but it was
> > not
> > > > > > > > > > released, and still took 100Gb, instead of expected 50Gb;
> > > > > > > > > >
> > > > > > > > > > Another use case:
> > > > > > > > > > - a user extracts data from IgniteCache to store it in
> > > separate
> > > > > > > > > > IgniteCache or another store;
> > > > > > > > > > - disc still is occupied and the user is not able to
> store
> > > data
> > > > > in the
> > > > > > > > > > different cache at the same cluster because of disc
> > > limitation;
> > > > > > > > > >
> > > > > > > > > > How can we help the user to free up the disc space, if an
> > > > amount
> > > > > of
> > > > > > > > > > data in IgniteCache has been reduced many times and will
> > not
> > > be
> > > > > > > > > > increased in the nearest future?
> > > > > > > > > >
> > > > > > > > > > AFAIK, we have mechanics of reusing memory pages, that
> > allows
> > > > us
> > > > > to
> > > > > > > > > > use pages which have been allocated and stored removed
> data
> > > for
> > > > > > > > > > storing new data.
> > > > > > > > > > Are there any chances to shrink data and free up space on
> > > disc
> > > > > (with
> > > > > > > > > > defragmentation if possible)?
> > > > > > > > > >
> > > > > > > > > > --
> > > > > > > > > > Best Regards, Vyacheslav D.
> > > > > > > > > >
> > > > > > > > >
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > --
> > > > > > > Best regards,
> > > > > > > Ivan Pavlukhin
> > > > > >
> > > > > >
> > > > > >
> > > > > > --
> > > > > > Best Regards, Vyacheslav D.
> > > > >
> > > > >
> > > > >
> > > > > --
> > > > > Best regards,
> > > > > Ivan Pavlukhin
> > > > >
> > > >
> > >
> >
>

[Alexey Goncharuk]

Anton,


> >>  The solution which Anton suggested does not look easy because it will
> most likely significantly hurt performance
> Mostly agree here, but what drop do we expect? What price do we ready to
> pay?
> Not sure, but seems some vendors ready to pay, for example, 5% drop for
> this.

5% may be a big drop for some use-cases, so I think we should look at how
to improve performance, not how to make it worse.

>
> >> it is hard to maintain a data structure to choose "page from free-list
> with enough space closest to the beginning of the file".
> We can just split each free-list bucket to the couple and use first for
> pages in the first half of the file and the second for the last.
> Only two buckets required here since, during the file shrink, first
> bucket's window will be shrank too.
> Seems, this give us the same price on put, just use the first bucket in
> case it's not empty.
> Remove price (with merge) will be increased, of course.
>
> The compromise solution is to have priority put (to the first path of the
> file), with keeping removal as is, and schedulable per-page migration for
> the rest of the data during the low activity period.
>

Free lists are large and slow by themselves, it is expensive to checkpoint
and read them on start, so as a long-term solution I would look into
removing them. Moreover, not sure if adding yet another background process
will improve the codebase reliability and simplicity.

If we want to go the hard path, I would look at free page tracking bitmap -
a special bitmask page, where each page in an adjacent block is marked as 0
if it has free space more than a certain configurable threshold (say, 80%)
- free, and 1 if less (full). Some vendors have successfully implemented
this approach, which looks much more promising, but harder to implement.

--AG

[Anton Vinogradov-2]

Alexey,

Let's combine your and Ivan's proposals.

>> vacuum command, which acquires exclusive table lock, so no concurrent
activities on the table are possible.
and
>> Could the problem be solved by stopping a node which needs to be
defragmented, clearing persistence files and restarting the node?
>> After rebalancing the node will receive all data back without
fragmentation.

How about to have special partition state SHRINKING?
This state should mean that partition unavailable for reads and updates but
should keep it's update-counters and should not be marked as lost, renting
or evicted.
At this state we able to iterate over the partition and apply it's entries
to another file in a compact way.
Indices should be updated during the copy-on-shrink procedure or at the
shrink completion.
Once shrank file is ready we should replace the original partition file
with it and mark it as MOVING which will start the historical rebalance.
Shrinking should be performed during the low activity periods, but even in
case we found that activity was high and historical rebalance is not
suitable we may just remove the file and use regular rebalance to restore
the partition (this will also lead to shrink).

BTW, seems, we able to implement partition shrink in a cheap way.
We may just use rebalancing code to apply fat partition's entries to the
new file.
So, 3 stages here: local rebalance, indices update and global historical
rebalance.

On Thu, Sep 19, 2019 at 11:43 AM Alexey Goncharuk <
[hidden email]> wrote:

> Anton,
>
>
> > >>  The solution which Anton suggested does not look easy because it will
> > most likely significantly hurt performance
> > Mostly agree here, but what drop do we expect? What price do we ready to
> > pay?
> > Not sure, but seems some vendors ready to pay, for example, 5% drop for
> > this.
>
> 5% may be a big drop for some use-cases, so I think we should look at how
> to improve performance, not how to make it worse.
>
>
> >
> > >> it is hard to maintain a data structure to choose "page from free-list
> > with enough space closest to the beginning of the file".
> > We can just split each free-list bucket to the couple and use first for
> > pages in the first half of the file and the second for the last.
> > Only two buckets required here since, during the file shrink, first
> > bucket's window will be shrank too.
> > Seems, this give us the same price on put, just use the first bucket in
> > case it's not empty.
> > Remove price (with merge) will be increased, of course.
> >
> > The compromise solution is to have priority put (to the first path of the
> > file), with keeping removal as is, and schedulable per-page migration for
> > the rest of the data during the low activity period.
> >
> Free lists are large and slow by themselves, it is expensive to checkpoint
> and read them on start, so as a long-term solution I would look into
> removing them. Moreover, not sure if adding yet another background process
> will improve the codebase reliability and simplicity.
>
> If we want to go the hard path, I would look at free page tracking bitmap -
> a special bitmask page, where each page in an adjacent block is marked as 0
> if it has free space more than a certain configurable threshold (say, 80%)
> - free, and 1 if less (full). Some vendors have successfully implemented
> this approach, which looks much more promising, but harder to implement.
>
> --AG
>

[Alexei Scherbakov]

The poor man's solution for the problem would be stopping fragmented node
and removing partition data, then starting it again allowing full state
transfer already without deletes.
Rinse and repeat for all owners.

Anton Vinogradov, would this work for you as workaround ?

чт, 19 сент. 2019 г. в 13:03, Anton Vinogradov <[hidden email]>:

> Alexey,
>
> Let's combine your and Ivan's proposals.
>
> >> vacuum command, which acquires exclusive table lock, so no concurrent
> activities on the table are possible.
> and
> >> Could the problem be solved by stopping a node which needs to be
> defragmented, clearing persistence files and restarting the node?
> >> After rebalancing the node will receive all data back without
> fragmentation.
>
> How about to have special partition state SHRINKING?
> This state should mean that partition unavailable for reads and updates but
> should keep it's update-counters and should not be marked as lost, renting
> or evicted.
> At this state we able to iterate over the partition and apply it's entries
> to another file in a compact way.
> Indices should be updated during the copy-on-shrink procedure or at the
> shrink completion.
> Once shrank file is ready we should replace the original partition file
> with it and mark it as MOVING which will start the historical rebalance.
> Shrinking should be performed during the low activity periods, but even in
> case we found that activity was high and historical rebalance is not
> suitable we may just remove the file and use regular rebalance to restore
> the partition (this will also lead to shrink).
>
> BTW, seems, we able to implement partition shrink in a cheap way.
> We may just use rebalancing code to apply fat partition's entries to the
> new file.
> So, 3 stages here: local rebalance, indices update and global historical
> rebalance.
>
> On Thu, Sep 19, 2019 at 11:43 AM Alexey Goncharuk <
> [hidden email]> wrote:
>
> > Anton,
> >
> >
> > > >>  The solution which Anton suggested does not look easy because it
> will
> > > most likely significantly hurt performance
> > > Mostly agree here, but what drop do we expect? What price do we ready
> to
> > > pay?
> > > Not sure, but seems some vendors ready to pay, for example, 5% drop for
> > > this.
> >
> > 5% may be a big drop for some use-cases, so I think we should look at how
> > to improve performance, not how to make it worse.
> >
> >
> > >
> > > >> it is hard to maintain a data structure to choose "page from
> free-list
> > > with enough space closest to the beginning of the file".
> > > We can just split each free-list bucket to the couple and use first for
> > > pages in the first half of the file and the second for the last.
> > > Only two buckets required here since, during the file shrink, first
> > > bucket's window will be shrank too.
> > > Seems, this give us the same price on put, just use the first bucket in
> > > case it's not empty.
> > > Remove price (with merge) will be increased, of course.
> > >
> > > The compromise solution is to have priority put (to the first path of
> the
> > > file), with keeping removal as is, and schedulable per-page migration
> for
> > > the rest of the data during the low activity period.
> > >
> > Free lists are large and slow by themselves, it is expensive to
> checkpoint
> > and read them on start, so as a long-term solution I would look into
> > removing them. Moreover, not sure if adding yet another background
> process
> > will improve the codebase reliability and simplicity.
> >
> > If we want to go the hard path, I would look at free page tracking
> bitmap -
> > a special bitmask page, where each page in an adjacent block is marked
> as 0
> > if it has free space more than a certain configurable threshold (say,
> 80%)
> > - free, and 1 if less (full). Some vendors have successfully implemented
> > this approach, which looks much more promising, but harder to implement.
> >
> > --AG
> >
>

--

Best regards,
Alexei Scherbakov

[Alexei Scherbakov]

Probably this should be allowed to do using public API, actually this is
same as manual rebalancing.

пт, 27 сент. 2019 г. в 17:40, Alexei Scherbakov <
[hidden email]>:

> The poor man's solution for the problem would be stopping fragmented node
> and removing partition data, then starting it again allowing full state
> transfer already without deletes.
> Rinse and repeat for all owners.
>
> Anton Vinogradov, would this work for you as workaround ?
>
> чт, 19 сент. 2019 г. в 13:03, Anton Vinogradov <[hidden email]>:
>
>> Alexey,
>>
>> Let's combine your and Ivan's proposals.
>>
>> >> vacuum command, which acquires exclusive table lock, so no concurrent
>> activities on the table are possible.
>> and
>> >> Could the problem be solved by stopping a node which needs to be
>> defragmented, clearing persistence files and restarting the node?
>> >> After rebalancing the node will receive all data back without
>> fragmentation.
>>
>> How about to have special partition state SHRINKING?
>> This state should mean that partition unavailable for reads and updates
>> but
>> should keep it's update-counters and should not be marked as lost, renting
>> or evicted.
>> At this state we able to iterate over the partition and apply it's entries
>> to another file in a compact way.
>> Indices should be updated during the copy-on-shrink procedure or at the
>> shrink completion.
>> Once shrank file is ready we should replace the original partition file
>> with it and mark it as MOVING which will start the historical rebalance.
>> Shrinking should be performed during the low activity periods, but even in
>> case we found that activity was high and historical rebalance is not
>> suitable we may just remove the file and use regular rebalance to restore
>> the partition (this will also lead to shrink).
>>
>> BTW, seems, we able to implement partition shrink in a cheap way.
>> We may just use rebalancing code to apply fat partition's entries to the
>> new file.
>> So, 3 stages here: local rebalance, indices update and global historical
>> rebalance.
>>
>> On Thu, Sep 19, 2019 at 11:43 AM Alexey Goncharuk <
>> [hidden email]> wrote:
>>
>> > Anton,
>> >
>> >
>> > > >>  The solution which Anton suggested does not look easy because it
>> will
>> > > most likely significantly hurt performance
>> > > Mostly agree here, but what drop do we expect? What price do we ready
>> to
>> > > pay?
>> > > Not sure, but seems some vendors ready to pay, for example, 5% drop
>> for
>> > > this.
>> >
>> > 5% may be a big drop for some use-cases, so I think we should look at
>> how
>> > to improve performance, not how to make it worse.
>> >
>> >
>> > >
>> > > >> it is hard to maintain a data structure to choose "page from
>> free-list
>> > > with enough space closest to the beginning of the file".
>> > > We can just split each free-list bucket to the couple and use first
>> for
>> > > pages in the first half of the file and the second for the last.
>> > > Only two buckets required here since, during the file shrink, first
>> > > bucket's window will be shrank too.
>> > > Seems, this give us the same price on put, just use the first bucket
>> in
>> > > case it's not empty.
>> > > Remove price (with merge) will be increased, of course.
>> > >
>> > > The compromise solution is to have priority put (to the first path of
>> the
>> > > file), with keeping removal as is, and schedulable per-page migration
>> for
>> > > the rest of the data during the low activity period.
>> > >
>> > Free lists are large and slow by themselves, it is expensive to
>> checkpoint
>> > and read them on start, so as a long-term solution I would look into
>> > removing them. Moreover, not sure if adding yet another background
>> process
>> > will improve the codebase reliability and simplicity.
>> >
>> > If we want to go the hard path, I would look at free page tracking
>> bitmap -
>> > a special bitmask page, where each page in an adjacent block is marked
>> as 0
>> > if it has free space more than a certain configurable threshold (say,
>> 80%)
>> > - free, and 1 if less (full). Some vendors have successfully implemented
>> > this approach, which looks much more promising, but harder to implement.
>> >
>> > --AG
>> >
>>
>
>
> --
>
> Best regards,
> Alexei Scherbakov
>

--

Best regards,
Alexei Scherbakov

[Anton Vinogradov-2]

Alexei,
>> stopping fragmented node and removing partition data, then starting it
again

That's exactly what we're doing to solve the fragmentation issue.
The problem here is that we have to perform N/B restart-rebalance
operations (N - cluster size, B - backups count) and it takes a lot of time
with risks to lose the data.

On Fri, Sep 27, 2019 at 5:49 PM Alexei Scherbakov <
[hidden email]> wrote:

> Probably this should be allowed to do using public API, actually this is
> same as manual rebalancing.
>
> пт, 27 сент. 2019 г. в 17:40, Alexei Scherbakov <
> [hidden email]>:
>
> > The poor man's solution for the problem would be stopping fragmented node
> > and removing partition data, then starting it again allowing full state
> > transfer already without deletes.
> > Rinse and repeat for all owners.
> >
> > Anton Vinogradov, would this work for you as workaround ?
> >
> > чт, 19 сент. 2019 г. в 13:03, Anton Vinogradov <[hidden email]>:
> >
> >> Alexey,
> >>
> >> Let's combine your and Ivan's proposals.
> >>
> >> >> vacuum command, which acquires exclusive table lock, so no concurrent
> >> activities on the table are possible.
> >> and
> >> >> Could the problem be solved by stopping a node which needs to be
> >> defragmented, clearing persistence files and restarting the node?
> >> >> After rebalancing the node will receive all data back without
> >> fragmentation.
> >>
> >> How about to have special partition state SHRINKING?
> >> This state should mean that partition unavailable for reads and updates
> >> but
> >> should keep it's update-counters and should not be marked as lost,
> renting
> >> or evicted.
> >> At this state we able to iterate over the partition and apply it's
> entries
> >> to another file in a compact way.
> >> Indices should be updated during the copy-on-shrink procedure or at the
> >> shrink completion.
> >> Once shrank file is ready we should replace the original partition file
> >> with it and mark it as MOVING which will start the historical rebalance.
> >> Shrinking should be performed during the low activity periods, but even
> in
> >> case we found that activity was high and historical rebalance is not
> >> suitable we may just remove the file and use regular rebalance to
> restore
> >> the partition (this will also lead to shrink).
> >>
> >> BTW, seems, we able to implement partition shrink in a cheap way.
> >> We may just use rebalancing code to apply fat partition's entries to the
> >> new file.
> >> So, 3 stages here: local rebalance, indices update and global historical
> >> rebalance.
> >>
> >> On Thu, Sep 19, 2019 at 11:43 AM Alexey Goncharuk <
> >> [hidden email]> wrote:
> >>
> >> > Anton,
> >> >
> >> >
> >> > > >>  The solution which Anton suggested does not look easy because it
> >> will
> >> > > most likely significantly hurt performance
> >> > > Mostly agree here, but what drop do we expect? What price do we
> ready
> >> to
> >> > > pay?
> >> > > Not sure, but seems some vendors ready to pay, for example, 5% drop
> >> for
> >> > > this.
> >> >
> >> > 5% may be a big drop for some use-cases, so I think we should look at
> >> how
> >> > to improve performance, not how to make it worse.
> >> >
> >> >
> >> > >
> >> > > >> it is hard to maintain a data structure to choose "page from
> >> free-list
> >> > > with enough space closest to the beginning of the file".
> >> > > We can just split each free-list bucket to the couple and use first
> >> for
> >> > > pages in the first half of the file and the second for the last.
> >> > > Only two buckets required here since, during the file shrink, first
> >> > > bucket's window will be shrank too.
> >> > > Seems, this give us the same price on put, just use the first bucket
> >> in
> >> > > case it's not empty.
> >> > > Remove price (with merge) will be increased, of course.
> >> > >
> >> > > The compromise solution is to have priority put (to the first path
> of
> >> the
> >> > > file), with keeping removal as is, and schedulable per-page
> migration
> >> for
> >> > > the rest of the data during the low activity period.
> >> > >
> >> > Free lists are large and slow by themselves, it is expensive to
> >> checkpoint
> >> > and read them on start, so as a long-term solution I would look into
> >> > removing them. Moreover, not sure if adding yet another background
> >> process
> >> > will improve the codebase reliability and simplicity.
> >> >
> >> > If we want to go the hard path, I would look at free page tracking
> >> bitmap -
> >> > a special bitmask page, where each page in an adjacent block is marked
> >> as 0
> >> > if it has free space more than a certain configurable threshold (say,
> >> 80%)
> >> > - free, and 1 if less (full). Some vendors have successfully
> implemented
> >> > this approach, which looks much more promising, but harder to
> implement.
> >> >
> >> > --AG
> >> >
> >>
> >
> >
> > --
> >
> > Best regards,
> > Alexei Scherbakov
> >
>
>
> --
>
> Best regards,
> Alexei Scherbakov
>

[Alexey Goncharuk]

Anton,

Switching a partition to and from the SHRINKING state will require
intricate synchronizations in order to properly determine the start
position for historical rebalance without PME.

I would still go with an offline-node approach, but instead of cleaning the
persistence, we can do effective defragmentation when the node is offline
because we are sure that there is no concurrent load. After the
defragmentation completes, we bring the node back to the cluster and
historical rebalance will kick in automatically. It will still require
manual node restarts, but since the data is not removed, there are no
additional risks. Also, this will be an excellent solution for those who
can afford downtime and execute the defragment command on all nodes in the
cluster simultaneously - this will be the fastest way possible.

--AG

пн, 30 сент. 2019 г. в 09:29, Anton Vinogradov <[hidden email]>:

> Alexei,
> >> stopping fragmented node and removing partition data, then starting it
> again
>
> That's exactly what we're doing to solve the fragmentation issue.
> The problem here is that we have to perform N/B restart-rebalance
> operations (N - cluster size, B - backups count) and it takes a lot of time
> with risks to lose the data.
>
> On Fri, Sep 27, 2019 at 5:49 PM Alexei Scherbakov <
> [hidden email]> wrote:
>
> > Probably this should be allowed to do using public API, actually this is
> > same as manual rebalancing.
> >
> > пт, 27 сент. 2019 г. в 17:40, Alexei Scherbakov <
> > [hidden email]>:
> >
> > > The poor man's solution for the problem would be stopping fragmented
> node
> > > and removing partition data, then starting it again allowing full state
> > > transfer already without deletes.
> > > Rinse and repeat for all owners.
> > >
> > > Anton Vinogradov, would this work for you as workaround ?
> > >
> > > чт, 19 сент. 2019 г. в 13:03, Anton Vinogradov <[hidden email]>:
> > >
> > >> Alexey,
> > >>
> > >> Let's combine your and Ivan's proposals.
> > >>
> > >> >> vacuum command, which acquires exclusive table lock, so no
> concurrent
> > >> activities on the table are possible.
> > >> and
> > >> >> Could the problem be solved by stopping a node which needs to be
> > >> defragmented, clearing persistence files and restarting the node?
> > >> >> After rebalancing the node will receive all data back without
> > >> fragmentation.
> > >>
> > >> How about to have special partition state SHRINKING?
> > >> This state should mean that partition unavailable for reads and
> updates
> > >> but
> > >> should keep it's update-counters and should not be marked as lost,
> > renting
> > >> or evicted.
> > >> At this state we able to iterate over the partition and apply it's
> > entries
> > >> to another file in a compact way.
> > >> Indices should be updated during the copy-on-shrink procedure or at
> the
> > >> shrink completion.
> > >> Once shrank file is ready we should replace the original partition
> file
> > >> with it and mark it as MOVING which will start the historical
> rebalance.
> > >> Shrinking should be performed during the low activity periods, but
> even
> > in
> > >> case we found that activity was high and historical rebalance is not
> > >> suitable we may just remove the file and use regular rebalance to
> > restore
> > >> the partition (this will also lead to shrink).
> > >>
> > >> BTW, seems, we able to implement partition shrink in a cheap way.
> > >> We may just use rebalancing code to apply fat partition's entries to
> the
> > >> new file.
> > >> So, 3 stages here: local rebalance, indices update and global
> historical
> > >> rebalance.
> > >>
> > >> On Thu, Sep 19, 2019 at 11:43 AM Alexey Goncharuk <
> > >> [hidden email]> wrote:
> > >>
> > >> > Anton,
> > >> >
> > >> >
> > >> > > >>  The solution which Anton suggested does not look easy because
> it
> > >> will
> > >> > > most likely significantly hurt performance
> > >> > > Mostly agree here, but what drop do we expect? What price do we
> > ready
> > >> to
> > >> > > pay?
> > >> > > Not sure, but seems some vendors ready to pay, for example, 5%
> drop
> > >> for
> > >> > > this.
> > >> >
> > >> > 5% may be a big drop for some use-cases, so I think we should look
> at
> > >> how
> > >> > to improve performance, not how to make it worse.
> > >> >
> > >> >
> > >> > >
> > >> > > >> it is hard to maintain a data structure to choose "page from
> > >> free-list
> > >> > > with enough space closest to the beginning of the file".
> > >> > > We can just split each free-list bucket to the couple and use
> first
> > >> for
> > >> > > pages in the first half of the file and the second for the last.
> > >> > > Only two buckets required here since, during the file shrink,
> first
> > >> > > bucket's window will be shrank too.
> > >> > > Seems, this give us the same price on put, just use the first
> bucket
> > >> in
> > >> > > case it's not empty.
> > >> > > Remove price (with merge) will be increased, of course.
> > >> > >
> > >> > > The compromise solution is to have priority put (to the first path
> > of
> > >> the
> > >> > > file), with keeping removal as is, and schedulable per-page
> > migration
> > >> for
> > >> > > the rest of the data during the low activity period.
> > >> > >
> > >> > Free lists are large and slow by themselves, it is expensive to
> > >> checkpoint
> > >> > and read them on start, so as a long-term solution I would look into
> > >> > removing them. Moreover, not sure if adding yet another background
> > >> process
> > >> > will improve the codebase reliability and simplicity.
> > >> >
> > >> > If we want to go the hard path, I would look at free page tracking
> > >> bitmap -
> > >> > a special bitmask page, where each page in an adjacent block is
> marked
> > >> as 0
> > >> > if it has free space more than a certain configurable threshold
> (say,
> > >> 80%)
> > >> > - free, and 1 if less (full). Some vendors have successfully
> > implemented
> > >> > this approach, which looks much more promising, but harder to
> > implement.
> > >> >
> > >> > --AG
> > >> >
> > >>
> > >
> > >
> > > --
> > >
> > > Best regards,
> > > Alexei Scherbakov
> > >
> >
> >
> > --
> >
> > Best regards,
> > Alexei Scherbakov
> >
>

[Anton Vinogradov-2]

Alexey,

Sounds good to me.

On Thu, Oct 3, 2019 at 10:51 AM Alexey Goncharuk <[hidden email]>
wrote:

> Anton,
>
> Switching a partition to and from the SHRINKING state will require
> intricate synchronizations in order to properly determine the start
> position for historical rebalance without PME.
>
> I would still go with an offline-node approach, but instead of cleaning the
> persistence, we can do effective defragmentation when the node is offline
> because we are sure that there is no concurrent load. After the
> defragmentation completes, we bring the node back to the cluster and
> historical rebalance will kick in automatically. It will still require
> manual node restarts, but since the data is not removed, there are no
> additional risks. Also, this will be an excellent solution for those who
> can afford downtime and execute the defragment command on all nodes in the
> cluster simultaneously - this will be the fastest way possible.
>
> --AG
>
> пн, 30 сент. 2019 г. в 09:29, Anton Vinogradov <[hidden email]>:
>
> > Alexei,
> > >> stopping fragmented node and removing partition data, then starting it
> > again
> >
> > That's exactly what we're doing to solve the fragmentation issue.
> > The problem here is that we have to perform N/B restart-rebalance
> > operations (N - cluster size, B - backups count) and it takes a lot of
> time
> > with risks to lose the data.
> >
> > On Fri, Sep 27, 2019 at 5:49 PM Alexei Scherbakov <
> > [hidden email]> wrote:
> >
> > > Probably this should be allowed to do using public API, actually this
> is
> > > same as manual rebalancing.
> > >
> > > пт, 27 сент. 2019 г. в 17:40, Alexei Scherbakov <
> > > [hidden email]>:
> > >
> > > > The poor man's solution for the problem would be stopping fragmented
> > node
> > > > and removing partition data, then starting it again allowing full
> state
> > > > transfer already without deletes.
> > > > Rinse and repeat for all owners.
> > > >
> > > > Anton Vinogradov, would this work for you as workaround ?
> > > >
> > > > чт, 19 сент. 2019 г. в 13:03, Anton Vinogradov <[hidden email]>:
> > > >
> > > >> Alexey,
> > > >>
> > > >> Let's combine your and Ivan's proposals.
> > > >>
> > > >> >> vacuum command, which acquires exclusive table lock, so no
> > concurrent
> > > >> activities on the table are possible.
> > > >> and
> > > >> >> Could the problem be solved by stopping a node which needs to be
> > > >> defragmented, clearing persistence files and restarting the node?
> > > >> >> After rebalancing the node will receive all data back without
> > > >> fragmentation.
> > > >>
> > > >> How about to have special partition state SHRINKING?
> > > >> This state should mean that partition unavailable for reads and
> > updates
> > > >> but
> > > >> should keep it's update-counters and should not be marked as lost,
> > > renting
> > > >> or evicted.
> > > >> At this state we able to iterate over the partition and apply it's
> > > entries
> > > >> to another file in a compact way.
> > > >> Indices should be updated during the copy-on-shrink procedure or at
> > the
> > > >> shrink completion.
> > > >> Once shrank file is ready we should replace the original partition
> > file
> > > >> with it and mark it as MOVING which will start the historical
> > rebalance.
> > > >> Shrinking should be performed during the low activity periods, but
> > even
> > > in
> > > >> case we found that activity was high and historical rebalance is not
> > > >> suitable we may just remove the file and use regular rebalance to
> > > restore
> > > >> the partition (this will also lead to shrink).
> > > >>
> > > >> BTW, seems, we able to implement partition shrink in a cheap way.
> > > >> We may just use rebalancing code to apply fat partition's entries to
> > the
> > > >> new file.
> > > >> So, 3 stages here: local rebalance, indices update and global
> > historical
> > > >> rebalance.
> > > >>
> > > >> On Thu, Sep 19, 2019 at 11:43 AM Alexey Goncharuk <
> > > >> [hidden email]> wrote:
> > > >>
> > > >> > Anton,
> > > >> >
> > > >> >
> > > >> > > >>  The solution which Anton suggested does not look easy
> because
> > it
> > > >> will
> > > >> > > most likely significantly hurt performance
> > > >> > > Mostly agree here, but what drop do we expect? What price do we
> > > ready
> > > >> to
> > > >> > > pay?
> > > >> > > Not sure, but seems some vendors ready to pay, for example, 5%
> > drop
> > > >> for
> > > >> > > this.
> > > >> >
> > > >> > 5% may be a big drop for some use-cases, so I think we should look
> > at
> > > >> how
> > > >> > to improve performance, not how to make it worse.
> > > >> >
> > > >> >
> > > >> > >
> > > >> > > >> it is hard to maintain a data structure to choose "page from
> > > >> free-list
> > > >> > > with enough space closest to the beginning of the file".
> > > >> > > We can just split each free-list bucket to the couple and use
> > first
> > > >> for
> > > >> > > pages in the first half of the file and the second for the last.
> > > >> > > Only two buckets required here since, during the file shrink,
> > first
> > > >> > > bucket's window will be shrank too.
> > > >> > > Seems, this give us the same price on put, just use the first
> > bucket
> > > >> in
> > > >> > > case it's not empty.
> > > >> > > Remove price (with merge) will be increased, of course.
> > > >> > >
> > > >> > > The compromise solution is to have priority put (to the first
> path
> > > of
> > > >> the
> > > >> > > file), with keeping removal as is, and schedulable per-page
> > > migration
> > > >> for
> > > >> > > the rest of the data during the low activity period.
> > > >> > >
> > > >> > Free lists are large and slow by themselves, it is expensive to
> > > >> checkpoint
> > > >> > and read them on start, so as a long-term solution I would look
> into
> > > >> > removing them. Moreover, not sure if adding yet another background
> > > >> process
> > > >> > will improve the codebase reliability and simplicity.
> > > >> >
> > > >> > If we want to go the hard path, I would look at free page tracking
> > > >> bitmap -
> > > >> > a special bitmask page, where each page in an adjacent block is
> > marked
> > > >> as 0
> > > >> > if it has free space more than a certain configurable threshold
> > (say,
> > > >> 80%)
> > > >> > - free, and 1 if less (full). Some vendors have successfully
> > > implemented
> > > >> > this approach, which looks much more promising, but harder to
> > > implement.
> > > >> >
> > > >> > --AG
> > > >> >
> > > >>
> > > >
> > > >
> > > > --
> > > >
> > > > Best regards,
> > > > Alexei Scherbakov
> > > >
> > >
> > >
> > > --
> > >
> > > Best regards,
> > > Alexei Scherbakov
> > >
> >
>

[Alexey Goncharuk]

Anton, do you think we should file a single ticket for this or should we go
with an IEP? As of now, the change does not look big enough for an IEP for
me.

чт, 3 окт. 2019 г. в 11:18, Anton Vinogradov <[hidden email]>:

> Alexey,
>
> Sounds good to me.
>
> On Thu, Oct 3, 2019 at 10:51 AM Alexey Goncharuk <
> [hidden email]>
> wrote:
>
> > Anton,
> >
> > Switching a partition to and from the SHRINKING state will require
> > intricate synchronizations in order to properly determine the start
> > position for historical rebalance without PME.
> >
> > I would still go with an offline-node approach, but instead of cleaning
> the
> > persistence, we can do effective defragmentation when the node is offline
> > because we are sure that there is no concurrent load. After the
> > defragmentation completes, we bring the node back to the cluster and
> > historical rebalance will kick in automatically. It will still require
> > manual node restarts, but since the data is not removed, there are no
> > additional risks. Also, this will be an excellent solution for those who
> > can afford downtime and execute the defragment command on all nodes in
> the
> > cluster simultaneously - this will be the fastest way possible.
> >
> > --AG
> >
> > пн, 30 сент. 2019 г. в 09:29, Anton Vinogradov <[hidden email]>:
> >
> > > Alexei,
> > > >> stopping fragmented node and removing partition data, then starting
> it
> > > again
> > >
> > > That's exactly what we're doing to solve the fragmentation issue.
> > > The problem here is that we have to perform N/B restart-rebalance
> > > operations (N - cluster size, B - backups count) and it takes a lot of
> > time
> > > with risks to lose the data.
> > >
> > > On Fri, Sep 27, 2019 at 5:49 PM Alexei Scherbakov <
> > > [hidden email]> wrote:
> > >
> > > > Probably this should be allowed to do using public API, actually this
> > is
> > > > same as manual rebalancing.
> > > >
> > > > пт, 27 сент. 2019 г. в 17:40, Alexei Scherbakov <
> > > > [hidden email]>:
> > > >
> > > > > The poor man's solution for the problem would be stopping
> fragmented
> > > node
> > > > > and removing partition data, then starting it again allowing full
> > state
> > > > > transfer already without deletes.
> > > > > Rinse and repeat for all owners.
> > > > >
> > > > > Anton Vinogradov, would this work for you as workaround ?
> > > > >
> > > > > чт, 19 сент. 2019 г. в 13:03, Anton Vinogradov <[hidden email]>:
> > > > >
> > > > >> Alexey,
> > > > >>
> > > > >> Let's combine your and Ivan's proposals.
> > > > >>
> > > > >> >> vacuum command, which acquires exclusive table lock, so no
> > > concurrent
> > > > >> activities on the table are possible.
> > > > >> and
> > > > >> >> Could the problem be solved by stopping a node which needs to
> be
> > > > >> defragmented, clearing persistence files and restarting the node?
> > > > >> >> After rebalancing the node will receive all data back without
> > > > >> fragmentation.
> > > > >>
> > > > >> How about to have special partition state SHRINKING?
> > > > >> This state should mean that partition unavailable for reads and
> > > updates
> > > > >> but
> > > > >> should keep it's update-counters and should not be marked as lost,
> > > > renting
> > > > >> or evicted.
> > > > >> At this state we able to iterate over the partition and apply it's
> > > > entries
> > > > >> to another file in a compact way.
> > > > >> Indices should be updated during the copy-on-shrink procedure or
> at
> > > the
> > > > >> shrink completion.
> > > > >> Once shrank file is ready we should replace the original partition
> > > file
> > > > >> with it and mark it as MOVING which will start the historical
> > > rebalance.
> > > > >> Shrinking should be performed during the low activity periods, but
> > > even
> > > > in
> > > > >> case we found that activity was high and historical rebalance is
> not
> > > > >> suitable we may just remove the file and use regular rebalance to
> > > > restore
> > > > >> the partition (this will also lead to shrink).
> > > > >>
> > > > >> BTW, seems, we able to implement partition shrink in a cheap way.
> > > > >> We may just use rebalancing code to apply fat partition's entries
> to
> > > the
> > > > >> new file.
> > > > >> So, 3 stages here: local rebalance, indices update and global
> > > historical
> > > > >> rebalance.
> > > > >>
> > > > >> On Thu, Sep 19, 2019 at 11:43 AM Alexey Goncharuk <
> > > > >> [hidden email]> wrote:
> > > > >>
> > > > >> > Anton,
> > > > >> >
> > > > >> >
> > > > >> > > >>  The solution which Anton suggested does not look easy
> > because
> > > it
> > > > >> will
> > > > >> > > most likely significantly hurt performance
> > > > >> > > Mostly agree here, but what drop do we expect? What price do
> we
> > > > ready
> > > > >> to
> > > > >> > > pay?
> > > > >> > > Not sure, but seems some vendors ready to pay, for example, 5%
> > > drop
> > > > >> for
> > > > >> > > this.
> > > > >> >
> > > > >> > 5% may be a big drop for some use-cases, so I think we should
> look
> > > at
> > > > >> how
> > > > >> > to improve performance, not how to make it worse.
> > > > >> >
> > > > >> >
> > > > >> > >
> > > > >> > > >> it is hard to maintain a data structure to choose "page
> from
> > > > >> free-list
> > > > >> > > with enough space closest to the beginning of the file".
> > > > >> > > We can just split each free-list bucket to the couple and use
> > > first
> > > > >> for
> > > > >> > > pages in the first half of the file and the second for the
> last.
> > > > >> > > Only two buckets required here since, during the file shrink,
> > > first
> > > > >> > > bucket's window will be shrank too.
> > > > >> > > Seems, this give us the same price on put, just use the first
> > > bucket
> > > > >> in
> > > > >> > > case it's not empty.
> > > > >> > > Remove price (with merge) will be increased, of course.
> > > > >> > >
> > > > >> > > The compromise solution is to have priority put (to the first
> > path
> > > > of
> > > > >> the
> > > > >> > > file), with keeping removal as is, and schedulable per-page
> > > > migration
> > > > >> for
> > > > >> > > the rest of the data during the low activity period.
> > > > >> > >
> > > > >> > Free lists are large and slow by themselves, it is expensive to
> > > > >> checkpoint
> > > > >> > and read them on start, so as a long-term solution I would look
> > into
> > > > >> > removing them. Moreover, not sure if adding yet another
> background
> > > > >> process
> > > > >> > will improve the codebase reliability and simplicity.
> > > > >> >
> > > > >> > If we want to go the hard path, I would look at free page
> tracking
> > > > >> bitmap -
> > > > >> > a special bitmask page, where each page in an adjacent block is
> > > marked
> > > > >> as 0
> > > > >> > if it has free space more than a certain configurable threshold
> > > (say,
> > > > >> 80%)
> > > > >> > - free, and 1 if less (full). Some vendors have successfully
> > > > implemented
> > > > >> > this approach, which looks much more promising, but harder to
> > > > implement.
> > > > >> >
> > > > >> > --AG
> > > > >> >
> > > > >>
> > > > >
> > > > >
> > > > > --
> > > > >
> > > > > Best regards,
> > > > > Alexei Scherbakov
> > > > >
> > > >
> > > >
> > > > --
> > > >
> > > > Best regards,
> > > > Alexei Scherbakov
> > > >
> > >
> >
>

[Anton Vinogradov-2]

Alexey,
As for me, it does not matter will it be IEP, umbrella or a single issue.
The most important thing is Assignee :)

On Thu, Oct 3, 2019 at 11:59 AM Alexey Goncharuk <[hidden email]>
wrote:

> Anton, do you think we should file a single ticket for this or should we go
> with an IEP? As of now, the change does not look big enough for an IEP for
> me.
>
> чт, 3 окт. 2019 г. в 11:18, Anton Vinogradov <[hidden email]>:
>
> > Alexey,
> >
> > Sounds good to me.
> >
> > On Thu, Oct 3, 2019 at 10:51 AM Alexey Goncharuk <
> > [hidden email]>
> > wrote:
> >
> > > Anton,
> > >
> > > Switching a partition to and from the SHRINKING state will require
> > > intricate synchronizations in order to properly determine the start
> > > position for historical rebalance without PME.
> > >
> > > I would still go with an offline-node approach, but instead of cleaning
> > the
> > > persistence, we can do effective defragmentation when the node is
> offline
> > > because we are sure that there is no concurrent load. After the
> > > defragmentation completes, we bring the node back to the cluster and
> > > historical rebalance will kick in automatically. It will still require
> > > manual node restarts, but since the data is not removed, there are no
> > > additional risks. Also, this will be an excellent solution for those
> who
> > > can afford downtime and execute the defragment command on all nodes in
> > the
> > > cluster simultaneously - this will be the fastest way possible.
> > >
> > > --AG
> > >
> > > пн, 30 сент. 2019 г. в 09:29, Anton Vinogradov <[hidden email]>:
> > >
> > > > Alexei,
> > > > >> stopping fragmented node and removing partition data, then
> starting
> > it
> > > > again
> > > >
> > > > That's exactly what we're doing to solve the fragmentation issue.
> > > > The problem here is that we have to perform N/B restart-rebalance
> > > > operations (N - cluster size, B - backups count) and it takes a lot
> of
> > > time
> > > > with risks to lose the data.
> > > >
> > > > On Fri, Sep 27, 2019 at 5:49 PM Alexei Scherbakov <
> > > > [hidden email]> wrote:
> > > >
> > > > > Probably this should be allowed to do using public API, actually
> this
> > > is
> > > > > same as manual rebalancing.
> > > > >
> > > > > пт, 27 сент. 2019 г. в 17:40, Alexei Scherbakov <
> > > > > [hidden email]>:
> > > > >
> > > > > > The poor man's solution for the problem would be stopping
> > fragmented
> > > > node
> > > > > > and removing partition data, then starting it again allowing full
> > > state
> > > > > > transfer already without deletes.
> > > > > > Rinse and repeat for all owners.
> > > > > >
> > > > > > Anton Vinogradov, would this work for you as workaround ?
> > > > > >
> > > > > > чт, 19 сент. 2019 г. в 13:03, Anton Vinogradov <[hidden email]>:
> > > > > >
> > > > > >> Alexey,
> > > > > >>
> > > > > >> Let's combine your and Ivan's proposals.
> > > > > >>
> > > > > >> >> vacuum command, which acquires exclusive table lock, so no
> > > > concurrent
> > > > > >> activities on the table are possible.
> > > > > >> and
> > > > > >> >> Could the problem be solved by stopping a node which needs to
> > be
> > > > > >> defragmented, clearing persistence files and restarting the
> node?
> > > > > >> >> After rebalancing the node will receive all data back without
> > > > > >> fragmentation.
> > > > > >>
> > > > > >> How about to have special partition state SHRINKING?
> > > > > >> This state should mean that partition unavailable for reads and
> > > > updates
> > > > > >> but
> > > > > >> should keep it's update-counters and should not be marked as
> lost,
> > > > > renting
> > > > > >> or evicted.
> > > > > >> At this state we able to iterate over the partition and apply
> it's
> > > > > entries
> > > > > >> to another file in a compact way.
> > > > > >> Indices should be updated during the copy-on-shrink procedure or
> > at
> > > > the
> > > > > >> shrink completion.
> > > > > >> Once shrank file is ready we should replace the original
> partition
> > > > file
> > > > > >> with it and mark it as MOVING which will start the historical
> > > > rebalance.
> > > > > >> Shrinking should be performed during the low activity periods,
> but
> > > > even
> > > > > in
> > > > > >> case we found that activity was high and historical rebalance is
> > not
> > > > > >> suitable we may just remove the file and use regular rebalance
> to
> > > > > restore
> > > > > >> the partition (this will also lead to shrink).
> > > > > >>
> > > > > >> BTW, seems, we able to implement partition shrink in a cheap
> way.
> > > > > >> We may just use rebalancing code to apply fat partition's
> entries
> > to
> > > > the
> > > > > >> new file.
> > > > > >> So, 3 stages here: local rebalance, indices update and global
> > > > historical
> > > > > >> rebalance.
> > > > > >>
> > > > > >> On Thu, Sep 19, 2019 at 11:43 AM Alexey Goncharuk <
> > > > > >> [hidden email]> wrote:
> > > > > >>
> > > > > >> > Anton,
> > > > > >> >
> > > > > >> >
> > > > > >> > > >>  The solution which Anton suggested does not look easy
> > > because
> > > > it
> > > > > >> will
> > > > > >> > > most likely significantly hurt performance
> > > > > >> > > Mostly agree here, but what drop do we expect? What price do
> > we
> > > > > ready
> > > > > >> to
> > > > > >> > > pay?
> > > > > >> > > Not sure, but seems some vendors ready to pay, for example,
> 5%
> > > > drop
> > > > > >> for
> > > > > >> > > this.
> > > > > >> >
> > > > > >> > 5% may be a big drop for some use-cases, so I think we should
> > look
> > > > at
> > > > > >> how
> > > > > >> > to improve performance, not how to make it worse.
> > > > > >> >
> > > > > >> >
> > > > > >> > >
> > > > > >> > > >> it is hard to maintain a data structure to choose "page
> > from
> > > > > >> free-list
> > > > > >> > > with enough space closest to the beginning of the file".
> > > > > >> > > We can just split each free-list bucket to the couple and
> use
> > > > first
> > > > > >> for
> > > > > >> > > pages in the first half of the file and the second for the
> > last.
> > > > > >> > > Only two buckets required here since, during the file
> shrink,
> > > > first
> > > > > >> > > bucket's window will be shrank too.
> > > > > >> > > Seems, this give us the same price on put, just use the
> first
> > > > bucket
> > > > > >> in
> > > > > >> > > case it's not empty.
> > > > > >> > > Remove price (with merge) will be increased, of course.
> > > > > >> > >
> > > > > >> > > The compromise solution is to have priority put (to the
> first
> > > path
> > > > > of
> > > > > >> the
> > > > > >> > > file), with keeping removal as is, and schedulable per-page
> > > > > migration
> > > > > >> for
> > > > > >> > > the rest of the data during the low activity period.
> > > > > >> > >
> > > > > >> > Free lists are large and slow by themselves, it is expensive
> to
> > > > > >> checkpoint
> > > > > >> > and read them on start, so as a long-term solution I would
> look
> > > into
> > > > > >> > removing them. Moreover, not sure if adding yet another
> > background
> > > > > >> process
> > > > > >> > will improve the codebase reliability and simplicity.
> > > > > >> >
> > > > > >> > If we want to go the hard path, I would look at free page
> > tracking
> > > > > >> bitmap -
> > > > > >> > a special bitmask page, where each page in an adjacent block
> is
> > > > marked
> > > > > >> as 0
> > > > > >> > if it has free space more than a certain configurable
> threshold
> > > > (say,
> > > > > >> 80%)
> > > > > >> > - free, and 1 if less (full). Some vendors have successfully
> > > > > implemented
> > > > > >> > this approach, which looks much more promising, but harder to
> > > > > implement.
> > > > > >> >
> > > > > >> > --AG
> > > > > >> >
> > > > > >>
> > > > > >
> > > > > >
> > > > > > --
> > > > > >
> > > > > > Best regards,
> > > > > > Alexei Scherbakov
> > > > > >
> > > > >
> > > > >
> > > > > --
> > > > >
> > > > > Best regards,
> > > > > Alexei Scherbakov
> > > > >
> > > >
> > >
> >
>