Gluster 9.0 - release roadmap

[msaju]

This issue is intended to receive updates for all the enhancements/improvements being proposed to be included in the release 9.0.

Release Deadlines:
Code Freeze : 20-Nov-2020
RC : 30-Nov-2020
GA : 11-Dec-2020

Planned Improvements for 9.0: (Additions on the way)

1. io_uring support for Gluster: This will be done in a phased manner.(https://review.gluster.org/#/c/glusterfs/+/24350/ to add io-uring support in posix xlator for reads, writes and fsyncs.)
2. Prevention of gfid split-brains
3. Read-dir performance improvement (https://review.gluster.org/24469, https://review.gluster.org/24470)
4. Compile with LTO/GCC 10
5. Monitor the performance in epoll/rpc layer (Currently, there is no way to find bottlenecks in rpc layer or monitor the performance/latency. This enhancement tries to fill that gap.)
6. glusterd-pmap binds to 10K ports on startup (using IPv4): glusterd-pmap binds to 10K ports on startup (using IPv4) #786
7. Change posix_threads(disk_space,health_check,fsync):In brick_mux environment currently we do spawn a separate posix_thread per brick, need to spwan a single thread per brick process
8. Implement brick resource cleanup code:Currently glusterd sends the kill signal to stop the brick without cleanup resources, Need to implement gracefull way to stop the brick
9. Glusterfind: internal DB :Glusterfind needs to improve the amount of time it takes processing the entries in its
   internal database when there's been data and metadata intensive workload.
10. Support for 5K volumes
11. Consistent cache: Currently, caching feature is spread among several xlators and there are known race conditions that make it necessary to disable them. With the consistent cache approach we intend to provide a performance improvement by serving more requests directly from a central cache xlator.
12. Changelog Enhancements: Enhancements include optimisation in changelog directory structure, enabling changelog enabled time (elimination of htime file), recording of fops from internal clients and using full path name for renaming.

Release Process:

1. Update the release document
2. Announce the release

[amarts]

I would request to add #280 for the list. The patches are out there, and only review pending.

[msaju]

Updates to the list:

13. Runtime adjustment of mount options / fuse tunables : Tunables of the fuse translator should be possible to change while the mount is in place.
14. Provide support for add, remove and replace brick for thin arbiter: Add, remove and replace brick volume operations functionality is missing currently

[mykaul]

Updates to the list:

1. Runtime adjustment of mount options / fuse tunables : Tunables of the fuse translator should be possible to change while the mount is in place.
2. Provide support for add, remove and replace brick for thin arbiter: Add, remove and replace brick volume operations functionality is missing currently

It's better to edit, if possible, the initial list, so it could be tracked in one place. Also, using Markdown, we can probably add checkboxes, so we can track what's been done and what isn't in yet.

[msaju]

Update on the Gluster Release 9.0 progress. Issue #280 has been added to the list.

• io_uring support for Gluster: IN PROGRESS (https://review.gluster.org/#/c/glusterfs/+/24350/)

• Prevention of gfid split-brains: IN PROGRESS(https://review.gluster.org/#/c/glusterfs/+/24373/)

• Read-dir performance improvement: IN PROGRESS (https://review.gluster.org/24469, https://review.gluster.org/24470)

• Compile with LTO/GCC 10: TBD

• Monitor the performance in epoll/rpc layer: COMPLETED (https://review.gluster.org/#/c/glusterfs/+/24955/)

• glusterd-pmap binds to 10K ports on startup (using IPv4): glusterd-pmap binds to 10K ports on startup (using IPv4) #786 :IN PROGRESS

• Change posix_threads(disk_space,health_check,fsync):IN PROGRESS

• Implement brick resource cleanup code: TBD

• Glusterfind: internal DB : Improve the internal DB processing : IN PROGRESS

• Support for 5K volumes: IN PROGRESS

• Consistent cache: central cache xlator: IN PROGRESS

• Changelog Enhancements: IN PROGRESS

• Runtime adjustment of mount options / fuse tunables : IN PROGRESS

• Provide support for add, remove and replace brick for thin arbiter: IN PROGRESS

• Use internal error codes instead of UNIX errnos Use internal error codes instead of UNIX errnos #280 : IN PROGRESS

[msaju]

Update on the Gluster Release 9.0 progress.

• io_uring support for Gluster: io_uring support in glusterfs main branch #1398 IN PROGRESS

• Rename Dir heal: AFR: Rename dir heal shouldn't delete the directory at oldlocation and recreate it at new-location #1211 COMPLETE

• Read-dir performance improvement: Slow readdir(p) performance in DHT #117 IN PROGRESS

• Compile with LTO/GCC 10: In Progress

• Monitor the performance in epoll/rpc layer: RPC handling latencies should be printed in statedump #1466 COMPLETE

• glusterd-pmap binds to 10K ports on startup (using IPv4): glusterd-pmap binds to 10K ports on startup (using IPv4) #786 COMPLETE

• [Brick-mux] Attach several posix threads with glusterfs_ctx: [Brick-mux] Attach several posix threads with glusterfs_ctx #1482 IN PROGRESS

• Implement brick resource cleanup code: TBD

• Glusterfind: internal DB : Improve the internal DB processing : glusterfind: Improve performance when few months of changes are requested at one shot #530 IN PROGRESS

• Support for 5K volumes: glusterd[brick_mux]: Optimize friend handshake code to avoid call_bail #1613 IN PROGRESS

• Consistent cache: central cache xlator: [RFE] Unified performance xlator #218 IN PROGRESS

• Changelog Enhancements: Optimized CHANGELOG #154 / Add identifier while recording internal traffic in Changelog #638 IN PROGRESS

• Runtime adjustment of mount options / fuse tunables : IN PROGRESS

• Provide support for add, remove and replace brick for thin arbiter: Add more volume operations for thin arbiter #1528 IN PROGRESS

• Use internal error codes instead of UNIX errnos Use internal error codes instead of UNIX errnos #280 : IN PROGRESS

[kalebskeithley]

wrt Compile with LTO/GCC 10: TBD

Gluster, as of 8.0, already compiles with gcc-10 and LTO on Fedora f32/f33/f34/rawhide, Ubuntu 20.10 (groovy), and OpenSUSE Tumbleweed.

This is a no-op at this point. The only changes required were the gfapi symbol versioning and commit 8eba9de

There is nothing to enable or configure. If the regression tests are running on fedora-32 or later (or one of those other distributions) then they are using gcc-10 and LTO.

[msaju]

wrt Compile with LTO/GCC 10: TBD

Gluster, as of 8.0, already compiles with gcc-10 and LTO on Fedora f32/f33/f34/rawhide, Ubuntu 20.10 (groovy), and OpenSUSE Tumbleweed.

This is a no-op at this point. The only changes required were the gfapi symbol versioning and commit 8eba9de

There is nothing to enable or configure. If the regression tests are running on fedora-32 or later (or one of those other distributions) then they are using gcc-10 and LTO.

We will be doing the regression and performance tests as part of this activity.

[pranithk]

@msaju Could you change the description of 1211 to rename dir heal instead of gfid split-brain prvention? This is phase-1 to achieve gfid-split-brain prevention

[mykaul]

#768 above is not complete, it's still open.

[tshacked]

There is nothing to enable or configure. If the regression tests are running on fedora-32 or later (or one of those other distributions) then they are using gcc-10 and LTO.

A tiny correction, on Fedora 32 there's a need to explicitly add the -flto option for enabling LTO optimization. @kalebskeithley, can you please confirm this?

This is my current steps on the "Compile with LTO/GCC 10" task, please let me know if anything is missing:

1. For checking performance benefit from LTO optimization I'm going to run gluster-performance-test-suite on two sets of gluster RPMs builds done on F32 using gcc 10.2.1, where for one build LTO got enabled.
2. according to (1) result we can consider adding LTO as a build configuration option (only for F32 machine?)

[kalebskeithley]

Fedora-32 and later has gcc-10. Fedora-33 is the first version that requires the use of LTO for all the various bundled packages, e.g. glusterfs, ceph, etc. By extension it is presumed that RHEL9 will require the use of LTO for downstream/product packages. (Also, gcc-11 is coming soon, in Fedora Rawhide/34 and Fedora Enterprise Linux Next (ELN) which will be the basis for RHEL9.)

If someone wants to enable LTO when they build on Fedora-32 and earlier and other distributions then -flto must be added to the build CFLAGS and LDFLAGS. LTO has been a gcc feature since at least gcc-4. (How well it works in versions before gcc-10 is unknown.)

When building RPMS on Fedora-33 and later -flto is already automatically added by rpmbuild. (But if you're just doing an ordinary ./configure && make on Fedora-33 and later you will have to add -flto to the flags.)

I am not aware of any pressing requirement to use LTO just because gcc-10 is being used. Who is asking for LTO all the time?

[msaju]

Status update on the Gluster 9.0 major release items as of today is as follows:

Development Complete: #1670 , #1398 , #1621 .

• io_uring support for Gluster: io_uring support in glusterfs main branch #1398 IN PROGRESS
• Rename Dir heal: AFR: Rename dir heal shouldn't delete the directory at oldlocation and recreate it at new-location #1211 COMPLETE
• Read-dir performance improvement: Slow readdir(p) performance in DHT #117 IN PROGRESS
• Compile with LTO/GCC 10: IN PROGRESS
• Monitor the performance in epoll/rpc layer: RPC handling latencies should be printed in statedump #1466 COMPLETE
• glusterd-pmap binds to 10K ports on startup (using IPv4): glusterd-pmap binds to 10K ports on startup (using IPv4) #786/glusterd: enhancement around port mapper #1670 IN PROGRESS
• [Brick-mux] Attach several posix threads with glusterfs_ctx: [Brick-mux] Attach several posix threads with glusterfs_ctx #1482 IN PROGRESS
• Implement brick resource cleanup code: TBD
• Glusterfind: internal DB : Improve the internal DB processing : glusterfind: Improve performance when few months of changes are requested at one shot #530 IN PROGRESS
• Support for 5K volumes: glusterd[brick_mux]: Optimize friend handshake code to avoid call_bail #1613 IN PROGRESS
• Consistent cache: central cache xlator: [RFE] Unified performance xlator #218 IN PROGRESS
• Changelog Enhancements: Optimized CHANGELOG #154 / Add identifier while recording internal traffic in Changelog #638 IN PROGRESS
• Runtime adjustment of mount options / fuse tunables : IN PROGRESS
• Provide support for add, remove and replace brick for thin arbiter: Add more volume operations for thin arbiter #1528 IN PROGRESS
• Use internal error codes instead of UNIX errnos Use internal error codes instead of UNIX errnos #280 : IN PROGRESS
• glusterd/afr: enable granular-entry-heal by default glusterd/afr: enable granular-entry-heal by default #1621: IN PROGRESS

[msaju]

Status update on the Gluster 9.0 major release items as of today is as follows:

• io_uring support for Gluster: io_uring support in glusterfs main branch #1398 COMPLETE
• Rename Dir heal: AFR: Rename dir heal shouldn't delete the directory at oldlocation and recreate it at new-location #1211 COMPLETE
• Read-dir performance improvement: Slow readdir(p) performance in DHT #117 IN PROGRESS
• Compile with LTO/GCC 10: IN PROGRESS
• Monitor the performance in epoll/rpc layer: RPC handling latencies should be printed in statedump #1466 COMPLETE
• glusterd-pmap binds to 10K ports on startup (using IPv4): glusterd-pmap binds to 10K ports on startup (using IPv4) #786/glusterd: enhancement around port mapper #1670 MERGE PENDING
• [Brick-mux] Attach several posix threads with glusterfs_ctx: [Brick-mux] Attach several posix threads with glusterfs_ctx #1482 IN PROGRESS
• Implement brick resource cleanup code: core: Implement gracefull shutdown for a brick process #1751/brick-process: call 'notify()' and 'fini()' of brick xlators in a proper cleanup (with or without brick-mux) #1749 IN PROGRESS
• Glusterfind: internal DB : Improve the internal DB processing : glusterfind: Improve performance when few months of changes are requested at one shot #530 IN PROGRESS
• Support for 5K volumes: glusterd[brick_mux]: Optimize friend handshake code to avoid call_bail #1613 IN PROGRESS
• Consistent cache: central cache xlator: [RFE] Unified performance xlator #218 IN PROGRESS
• Changelog Enhancements: Optimized CHANGELOG #154 / Add identifier while recording internal traffic in Changelog #638 IN PROGRESS
• Runtime adjustment of mount options / fuse tunables : IN PROGRESS
• Provide support for add, remove and replace brick for thin arbiter: Add more volume operations for thin arbiter #1528 IN PROGRESS
• Use internal error codes instead of UNIX errnos Use internal error codes instead of UNIX errnos #280 : IN PROGRESS
• glusterd/afr: enable granular-entry-heal by default glusterd/afr: enable granular-entry-heal by default #1621: IN PROGRESS

[msaju]

Status update on the Gluster 9.0 major release items as of today is as follows:

• io_uring support for Gluster: io_uring support in glusterfs main branch #1398 COMPLETE
• Rename Dir heal: AFR: Rename dir heal shouldn't delete the directory at oldlocation and recreate it at new-location #1211 COMPLETE
• Read-dir performance improvement: Slow readdir(p) performance in DHT #117 TENTATIVE
• Compile with LTO/GCC 10: IN PROGRESS
• Monitor the performance in epoll/rpc layer: RPC handling latencies should be printed in statedump #1466 COMPLETE
• glusterd-pmap binds to 10K ports on startup (using IPv4): glusterd-pmap binds to 10K ports on startup (using IPv4) #786/glusterd: enhancement around port mapper #1670 MERGE PENDING
• [Brick-mux] Attach several posix threads with glusterfs_ctx: [Brick-mux] Attach several posix threads with glusterfs_ctx #1482 COMPLETE
• Implement brick resource cleanup code: core: Implement gracefull shutdown for a brick process #1751/brick-process: call 'notify()' and 'fini()' of brick xlators in a proper cleanup (with or without brick-mux) #1749 IN PROGRESS
• Glusterfind: internal DB : Improve the internal DB processing : glusterfind: Improve performance when few months of changes are requested at one shot #530 IN PROGRESS
• Support for 5K volumes: glusterd[brick_mux]: Optimize friend handshake code to avoid call_bail #1613 IN PROGRESS
• Consistent cache: central cache xlator: [RFE] Unified performance xlator #218 IN PROGRESS
• Changelog Enhancements: Optimized CHANGELOG #154 / Add identifier while recording internal traffic in Changelog #638 IN PROGRESS
• Runtime adjustment of mount options / fuse tunables : IN PROGRESS
• Provide support for add, remove and replace brick for thin arbiter: Add more volume operations for thin arbiter #1528 IN PROGRESS
• Use internal error codes instead of UNIX errnos Use internal error codes instead of UNIX errnos #280 : IN PROGRESS
• glusterd/afr: enable granular-entry-heal by default Enable granular-entry-heal by default #1483: COMPLETE

[msaju]

Status update on the Gluster 9.0 major release items as of today is as follows.
Few of the inital planned items are not ready for release timeline.
Those marked NOT READY FOR RELEASE, will be planned to be part of subsequent minor releases.
The GA date also has been re planned to be on 18th December 2020.

• io_uring support for Gluster: io_uring support in glusterfs main branch #1398 COMPLETE
• Rename Dir heal: AFR: Rename dir heal shouldn't delete the directory at oldlocation and recreate it at new-location #1211 COMPLETE
• Read-dir performance improvement: Slow readdir(p) performance in DHT #117 NOT READY FOR RELEASE
• Compile with LTO/GCC 10: build: add LTO as a configure option #1772 IN PROGRESS
• Monitor the performance in epoll/rpc layer: RPC handling latencies should be printed in statedump #1466 COMPLETE
• glusterd-pmap binds to 10K ports on startup (using IPv4): glusterd-pmap binds to 10K ports on startup (using IPv4) #786/glusterd: enhancement around port mapper #1670 MERGE PENDING
• [Brick-mux] Attach several posix threads with glusterfs_ctx: [Brick-mux] Attach several posix threads with glusterfs_ctx #1482 COMPLETE
• Implement brick resource cleanup code: core: Implement gracefull shutdown for a brick process #1751/brick-process: call 'notify()' and 'fini()' of brick xlators in a proper cleanup (with or without brick-mux) #1749 NOT READY FOR RELEASE
• Glusterfind: internal DB : Improve the internal DB processing : glusterfind: Improve performance when few months of changes are requested at one shot #530 NOT READY FOR RELEASE
• Support for 5K volumes: glusterd[brick_mux]: Optimize friend handshake code to avoid call_bail #1613 COMPLETE
• Consistent cache: central cache xlator: [RFE] Unified performance xlator #218 NOT READY FOR RELEASE
• Changelog Enhancements: Optimized CHANGELOG #154 / Add identifier while recording internal traffic in Changelog #638 IN PROGRESS
• Runtime adjustment of mount options / fuse tunables : NOT READY FOR RELEASE
• Provide support for add, remove and replace brick for thin arbiter: Add more volume operations for thin arbiter #1528 IN PROGRESS
• Use internal error codes instead of UNIX errnos Use internal error codes instead of UNIX errnos #280 : NOT READY FOR RELEASE
• glusterd/afr: enable granular-entry-heal by default Enable granular-entry-heal by default #1483: COMPLETE

[msaju]

Release 9.0 has been completed.