Some of our pilorama tests fail on CI.
The reasons are not obvious, but one possible improvement
is using `WithNoSync` option for these. It should have much effect,
because we are writing on the tmpfs, but doesn't hurt anyway.
If I replace `t.TempDir()` with a local directory, test execution time
goes down from 5s (sync) to 0.4s (nosync), which is the same time as
with `t.TempDir()`. Maybe we have some strange CI configuration.
```
panic: test timed out after 10m0s
running tests:
TestForest_ApplyRandom (8m22s)
TestForest_ApplyRandom/bbolt (8m21s)
...
goroutine 170 [syscall]:
syscall.Syscall(0xc000100000?, 0xc00047b758?, 0x6aff9a?, 0xc00041c1b0?)
/opt/hostedtoolcache/go/1.20.7/x64/src/syscall/syscall_linux.go:69 +0x27
syscall.Fdatasync(0x9e35c0?)
/opt/hostedtoolcache/go/1.20.7/x64/src/syscall/zsyscall_linux_amd64.go:418 +0x2a
go.etcd.io/bbolt.fdatasync(0xc000189000?)
```
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
The command is used in multiple places across the whole FrostFS
ecosystem. While we want to have uniform interfaces everywhere,
sometimes we can't: already defined global flags can be harder to change
because of our obligations to the users. Cobra framework doesn't allow
conflicting flags (we can have global ones), so allow to override them.
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
Make it harder to encounter conflicts in already existing commands.
Because the command is executed once, I don't think the usability is
worse.
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
Once the node was processed it skipped, at the step of forming
result in case when all nodes skipped, because processed for
previous REP, service mark the whole request as incomplete.
Example of policies which are unblocked:
- REP 1 REP 1 CBF 1
- REP 4 IN X REP 4 IN Y
CBF 4
SELECT 2 FROM FX AS X SELECT 2 FROM FY AS Y
FILTER Country EQ Russia OR Country EQ Sweden OR Country EQ Finland AS FY
FILTER Price GE 0 AS FX
Signed-off-by: Anton Nikiforov <an.nikiforov@yadro.com>
Estimate cache size after delete objects to update metric.
Update counters on small object deletion.
Do not count bbolt DB file as FSTree object.
Signed-off-by: Dmitrii Stepanov <d.stepanov@yadro.com>
We do not use the return result from Close() and we always execute both
methods in succession. It makes sense to unite them.
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
Current implementation has some quirks. For example,
using only half of object.put.pool_size_remote threads
tells replicator that is node is 50% loaded,
but in reality we could be putting lot's of big objects.
Signed-off-by: Dmitrii Stepanov <d.stepanov@yadro.com>
