TrueCloudLab/frostfs-node

Fork 27

WIP: Put a tombstone on a different shard if a target shard is read-only #1403

Closed

a-savchuk wants to merge 3 commits from a-savchuk/frostfs-node:put-tombstone-on-different-shard into master

a-savchuk commented

2024-10-01 09:44:17 +00:00

Member

Currently, when an object on some shard is inhumed, the engine places a tombstone on the same shard. If the target shard is read-only, the engine fails.

In that case, we want the engine to correctly place a tombstone on a different shard, ensuring that:

An object becomes unavailable if a tombstone was placed on a different shard.
GC deletes an object if a tombstone was placed on a different shard.

Signed-off-by: Aleksey Savchuk a.savchuk@yadro.com

Currently, when an object on some shard is inhumed, the engine places a tombstone on the same shard. If the target shard is read-only, the engine fails. In that case, we want the engine to correctly place a tombstone on a different shard, ensuring that: - [x] An object becomes unavailable if a tombstone was placed on a different shard. - [ ] GC deletes an object if a tombstone was placed on a different shard. Signed-off-by: Aleksey Savchuk <a.savchuk@yadro.com>

a-savchuk added 1 commit 2024-10-01 09:44:18 +00:00

[#xx] engine: Add tests for object deletion on read-only shard

DCO action / DCO (pull_request) Failing after 1m9s

Details

Tests and linters / Run gofumpt (pull_request) Successful in 2m26s

Details

Vulncheck / Vulncheck (pull_request) Successful in 2m35s

Details

Pre-commit hooks / Pre-commit (pull_request) Successful in 2m41s

Details

Build / Build Components (pull_request) Successful in 3m6s

Details

Tests and linters / Lint (pull_request) Successful in 3m47s

Details

Tests and linters / Staticcheck (pull_request) Successful in 3m46s

Details

Tests and linters / gopls check (pull_request) Successful in 3m48s

Details

Tests and linters / Tests (pull_request) Failing after 6m8s

Details

Tests and linters / Tests with -race (pull_request) Failing after 6m11s

Details

0c15185883

Currently, when an object on some shard is inhumed, the engine places a
tombstone on the same shard. If the target shard is read-only, the engine fails.

In that case, we want the engine to correctly place a tombstone on a different
shard, ensuring that:
- An object becomes unavailable if a tombstone was placed on a different shard.
  See `TestObjectUnavailableIfTombstoneOnDifferentShard` test.
- GC deletes an object if a tombstone was placed on a different shard.
  See `TestObjectDeletedIfTombstoneOnDifferentShard` test.

Signed-off-by: Aleksey Savchuk <a.savchuk@yadro.com>

a-savchuk force-pushed put-tombstone-on-different-shard from 0c15185883 to d511a08b60

2024-10-01 09:45:26 +00:00

Compare

a-savchuk commented

2024-10-01 09:48:23 +00:00

Author

Member

Both tests should fail before the bug is fixed

a-savchuk force-pushed put-tombstone-on-different-shard from d511a08b60 to 13f66540c1

2024-10-07 12:45:18 +00:00

Compare

a-savchuk force-pushed put-tombstone-on-different-shard from 13f66540c1 to 4d79969d12

2024-10-07 12:46:23 +00:00

Compare

a-savchuk force-pushed put-tombstone-on-different-shard from 4d79969d12 to f3735d0fb9

2024-10-07 12:56:47 +00:00

Compare

a-savchuk force-pushed put-tombstone-on-different-shard from f3735d0fb9 to 3c1aaee84f

2024-10-07 13:09:23 +00:00

Compare

a-savchuk force-pushed put-tombstone-on-different-shard from 3c1aaee84f to c4fca96dd1

2024-10-07 13:20:53 +00:00

Compare

a-savchuk force-pushed put-tombstone-on-different-shard from c4fca96dd1 to 2dc934839a

2024-10-08 09:29:39 +00:00

Compare

a-savchuk commented

2024-10-14 08:35:34 +00:00

Author

Member

We decided that supporting the ability to write a GC mark to a different shard leads to

A significant slowdown in read operations
- If an object and its mark are on different shards, all the shards must be checked to ensure this
- The situation where an object and its label are on different shards is rare, but checking all the shards will be necessary every time
- The number of shards can be quite large (~100-200 disks)
Increased complexity for the GC

I close this PR

We decided that supporting the ability to write a GC mark to a different shard leads to - A significant slowdown in read operations - If an object and its mark are on different shards, all the shards must be checked to ensure this - The situation where an object and its label are on different shards is rare, but checking all the shards will be necessary every time - The number of shards can be quite large (~100-200 disks) - Increased complexity for the GC I close this PR