Repository navigation
Conversation
|
Note This is an automated comment that will be appended during run. Note All workloads for linux-x86_64-relwithdebinfo have completed. Tip Planned checks for linux-x86_64-relwithdebinfo.
🟢 linux-x86_64-relwithdebinfo target: cloud/tasks/,cloud/storage/,cloud/fastshard/ (test time: 278s): all tests PASSED for commit 6a81a1f.
🟢 linux-x86_64-relwithdebinfo target: cloud/disk_manager/ (test time: 323s): all tests PASSED for commit 6a81a1f.
|
|
Codex review: • Reviewed 6a81a1f (validate gc in loadtests). The main concerns are the changed forced-flush contract and incomplete coverage of “no garbage left.” Newly introduced issues / behavior changes
Pre-existing harness issues — not regressions from this commit • Session setup failure can strand the controller. In test.cpp:371 (/home/anton-dmitrovsky/nbs/cloud/filestore/tools/testing/loadtest/lib/test.cpp:371), WaitForCompletion throws before cmd.Complete is fulfilled. The worker’s catch publishes its result and queues TTestFinished, while the controller is blocked waiting on the separate setup promise. That • Exceptional worker exit does not ensure teardown. The catch at test.cpp:497 (/home/anton-dmitrovsky/nbs/cloud/filestore/tools/testing/loadtest/lib/test.cpp:497) exits without draining requests, destroying the session, or stopping clients. Enqueuing teardown afterward cannot make the exited thread execute it. Consequently, the new validator’s assumption that • Existing shutdown waits remain unbounded. test.cpp:790 (/home/anton-dmitrovsky/nbs/cloud/filestore/tools/testing/loadtest/lib/test.cpp:790) drains I/O without a deadline; existing session/filesystem waits also retain the infinite default. The new GC timeout bounds only the new validation phase. C++ and simplification suggestions • Make ValidateStorageStats a free helper or static member: it does not use controller state. I found no obvious new C++ lifetime issue. Uncached statistics correctly aggregate shards, and normal teardown does drain outstanding I/O. This was a source review; git diff --check passed, but I did not build or run tests. |
|
Note This is an automated comment that will be appended during run. Note All workloads for linux-x86_64-relwithdebinfo have completed. Tip Planned checks for linux-x86_64-relwithdebinfo.
|
|
Note This is an automated comment that will be appended during run. Note All workloads for linux-x86_64-relwithdebinfo have completed. Tip Planned checks for linux-x86_64-relwithdebinfo. 🟢 linux-x86_64-relwithdebinfo target: cloud/tasks/,cloud/storage/,cloud/fastshard/ (test time: 314s): all tests PASSED for commit 5ac84d4.
🟢 linux-x86_64-relwithdebinfo target: cloud/disk_manager/ (test time: 2165s): all tests PASSED for commit 5ac84d4.
🟢 linux-x86_64-relwithdebinfo target: cloud/filestore/ (test time: 7009s): all tests PASSED for commit 5ac84d4.
🟢 linux-x86_64-relwithdebinfo target: cloud/blockstore/ (test time: 11499s): all tests PASSED for commit 5ac84d4.
|
|
Note This is an automated comment that will be appended during run. Note All workloads for linux-x86_64-relwithdebinfo have completed. Tip Planned checks for linux-x86_64-relwithdebinfo.
|
|
Note This is an automated comment that will be appended during run. Important Workload for linux-x86_64-relwithdebinfo is not finished yet. This comment will be updated after all workloads complete. Tip Planned checks for linux-x86_64-relwithdebinfo. |
Notes
This is the final PR enabling garbage collection validation in selected load tests.
When enabled in loadtest config, the test collects filesystem topology and issues flush, flush_bytes, cleanup, compaction, and collect_garbage commands to all fs shards. After cleanup, the test validates that certain TStorageStats counters are zero.
It includes replacing potentially unbounded forced flush_bytes operation with repeated calls from the test runner.
Issue
#4115