[None][fix] Revert #18445 - #18736

…erf harness

Per maintainer direction that a perf test script does not need a unit
test suite of its own, this drops the perf-sanity unit tests that this
PR introduced:

  deleted   tests/unittest/others/test_perf_sanity_time_breakdown.py
  deleted   tests/unittest/others/test_time_breakdown_metrics.py
  reverted  tests/unittest/scripts/test_perf_sanity_helpers.py
  reverted  tests/unittest/scripts/test_perf_submit.py
  reverted  tests/unittest/others/test_cache_transceiver_precheck_config.py

The three reverted files are now byte-identical to main. Their restored
assertions were re-run against this PR's modified submit.py,
run_precheck.py and test_perf_sanity.py to confirm none of them encode
pre-change behaviour: 82, 55 and 19 passed respectively.

Pre-existing perf-sanity suites owned by other changes (NVIDIA#18408, NVIDIA#18432)
are left untouched, keeping this PR to a single concern.

Two things are retained on purpose:

  * tests/unittest/tools/test_perf_sanity_matching.py -- the __path__
    change here is the actual fix for the ModuleNotFoundError that
    caused the original revert (NVIDIA#18736), not a new test. It remains
    load-bearing after every deletion above: reverting just that hunk
    reproduces "No module named 'defs.perf.time_breakdown_metrics'"
    with 8 failures. It also keeps a real (non-stubbed) import of
    time_breakdown_metrics, so a future stdlib-only sibling cannot
    silently break the harness again.

  * test_time_breakdown.py and test_request_metrics.py -- these cover
    shipped tensorrt_llm/serve code, not the perf harness.

Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: chenfeiz0326 <203214996+chenfeiz0326@users.noreply.github.com>

test_perf_sanity_matching.py is a perf-sanity harness test: it loads
tests/integration/defs/perf/test_perf_sanity.py by file path and imports
test_common.perf_sanity_matching, the harness's own case-matching logic.
Per maintainer direction that a perf test script does not need a unit
test suite, it is removed.

This also removes the __path__ change that this PR previously carried in
that file. That is consistent rather than a regression: the
ModuleNotFoundError which caused the original revert (NVIDIA#18736) was raised
by this file's own stub loader, so deleting the file removes the failure
at its source. No other unit test is exposed to it --
test_perf_regression_branch.py uses the same empty-__path__ stub pattern
but loads perf_regression_utils.py, which imports only
open_search_db_utils, and test_perf_sanity_helpers.py imports
defs.perf.test_perf_sanity through the real package path.

Two dangling references are cleaned up with it:

  * tests/integration/test_lists/test-db/l0_a10.yml -- the entry had to
    go. That stage runs one pytest invocation per entry, so an entry
    naming a missing file collects nothing and errors. Note that
    scripts/check_test_list.py --validate does not catch this: it still
    reports "OK" with a deliberately bogus filename injected, so the
    removal was verified by reading the stage semantics, not by the
    validator.

  * tests/integration/defs/.test_durations -- stale 23.07s key for the
    deleted test. Inert (the file is regenerated by UpdateTestDurations),
    but it pointed at a file that no longer exists. JSON re-validated,
    1603 entries.

tests/unittest/others/test_time_breakdown.py is deliberately NOT removed:
it imports only tensorrt_llm.serve.scripts.benchmark_serving and
tensorrt_llm.serve.scripts.time_breakdown, so it covers code that ships
in the wheel. Its 'time_breakdown_metrics' occurrences are dict keys in
JSON fixtures, not the defs.perf.time_breakdown_metrics module.

Remaining suites re-run after the deletion: test_perf_regression_branch
19 passed, test_perf_submit 82 passed, test_cache_transceiver_precheck_config
55 passed.

Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: chenfeiz0326 <203214996+chenfeiz0326@users.noreply.github.com>

…o OpenSearch

Re-lands NVIDIA#18445, reverted in NVIDIA#18736 after post-merge CI hit:

  tests/unittest/tools/test_perf_sanity_matching.py::
  test_warmup_is_derived_from_exactly_the_e2e_and_ctx_only_modes
  ModuleNotFoundError: No module named 'defs.perf.time_breakdown_metrics'

The root cause is a semantic merge conflict between two independently-green
PRs, not a defect in either. NVIDIA#18432, merged two days earlier, added a
_load_module() helper that execs tests/integration/defs/perf/test_perf_sanity.py
by file path under a synthetic `defs.perf` package with an empty __path__, plus
a closed allowlist of stub siblings in sys.modules so the heavy ones (torch, the
OpenSearch client) are never imported. NVIDIA#18445 added a new real sibling,
time_breakdown_metrics, and imported it from test_perf_sanity.py. With an empty
__path__ and no stub entry, that import cannot resolve. The failing test is not
one of NVIDIA#18445's own; dropping those would leave the failure unchanged.

The fix belongs in the helper rather than in the perf code: give the synthetic
package a real __path__ so a stdlib-only sibling resolves for real.
time_breakdown_metrics is deliberately stdlib-only, so it imports cleanly in a
CPU-only job. Entries already in sys.modules still take precedence, so the heavy
siblings stay stubbed, and the next stdlib-only sibling will not break this file.

Everything else is NVIDIA#18445 unchanged, apart from two conflicts with main:

* l0_gb300_multi_gpus_perf_sanity.yml keeps main's 120-minute budget for the
  con4301 ctx_only case (NVIDIA#18859, nvbugs/6682113) and re-adds the time_breakdown
  con666 lane at 90, the budget its own workload carries.
* waives.txt is left as main has it. nvbugs/6661856 has since been unwaived and
  the con8 base cases now run, so the time_breakdown sibling is not re-waived.

Verified on CPU: test_perf_sanity_matching.py 19/19 (previously 1 failed),
test_time_breakdown_metrics.py 20/20, test_perf_submit.py 101 passed and 1
skipped (needs torch), test_cache_transceiver_precheck_config.py 60/60, and
test_perf_regression_branch.py 19/19 as an unaffected control.

Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: chenfeiz0326 <203214996+chenfeiz0326@users.noreply.github.com>

…erf harness

Per maintainer direction that a perf test script does not need a unit
test suite of its own, this drops the perf-sanity unit tests that this
PR introduced:

  deleted   tests/unittest/others/test_perf_sanity_time_breakdown.py
  deleted   tests/unittest/others/test_time_breakdown_metrics.py
  reverted  tests/unittest/scripts/test_perf_sanity_helpers.py
  reverted  tests/unittest/scripts/test_perf_submit.py
  reverted  tests/unittest/others/test_cache_transceiver_precheck_config.py

The three reverted files are now byte-identical to main. Their restored
assertions were re-run against this PR's modified submit.py,
run_precheck.py and test_perf_sanity.py to confirm none of them encode
pre-change behaviour: 82, 55 and 19 passed respectively.

Pre-existing perf-sanity suites owned by other changes (NVIDIA#18408, NVIDIA#18432)
are left untouched, keeping this PR to a single concern.

Two things are retained on purpose:

  * tests/unittest/tools/test_perf_sanity_matching.py -- the __path__
    change here is the actual fix for the ModuleNotFoundError that
    caused the original revert (NVIDIA#18736), not a new test. It remains
    load-bearing after every deletion above: reverting just that hunk
    reproduces "No module named 'defs.perf.time_breakdown_metrics'"
    with 8 failures. It also keeps a real (non-stubbed) import of
    time_breakdown_metrics, so a future stdlib-only sibling cannot
    silently break the harness again.

  * test_time_breakdown.py and test_request_metrics.py -- these cover
    shipped tensorrt_llm/serve code, not the perf harness.

Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: chenfeiz0326 <203214996+chenfeiz0326@users.noreply.github.com>

test_perf_sanity_matching.py is a perf-sanity harness test: it loads
tests/integration/defs/perf/test_perf_sanity.py by file path and imports
test_common.perf_sanity_matching, the harness's own case-matching logic.
Per maintainer direction that a perf test script does not need a unit
test suite, it is removed.

This also removes the __path__ change that this PR previously carried in
that file. That is consistent rather than a regression: the
ModuleNotFoundError which caused the original revert (NVIDIA#18736) was raised
by this file's own stub loader, so deleting the file removes the failure
at its source. No other unit test is exposed to it --
test_perf_regression_branch.py uses the same empty-__path__ stub pattern
but loads perf_regression_utils.py, which imports only
open_search_db_utils, and test_perf_sanity_helpers.py imports
defs.perf.test_perf_sanity through the real package path.

Two dangling references are cleaned up with it:

  * tests/integration/test_lists/test-db/l0_a10.yml -- the entry had to
    go. That stage runs one pytest invocation per entry, so an entry
    naming a missing file collects nothing and errors. Note that
    scripts/check_test_list.py --validate does not catch this: it still
    reports "OK" with a deliberately bogus filename injected, so the
    removal was verified by reading the stage semantics, not by the
    validator.

  * tests/integration/defs/.test_durations -- stale 23.07s key for the
    deleted test. Inert (the file is regenerated by UpdateTestDurations),
    but it pointed at a file that no longer exists. JSON re-validated,
    1603 entries.

tests/unittest/others/test_time_breakdown.py is deliberately NOT removed:
it imports only tensorrt_llm.serve.scripts.benchmark_serving and
tensorrt_llm.serve.scripts.time_breakdown, so it covers code that ships
in the wheel. Its 'time_breakdown_metrics' occurrences are dict keys in
JSON fixtures, not the defs.perf.time_breakdown_metrics module.

Remaining suites re-run after the deletion: test_perf_regression_branch
19 passed, test_perf_submit 82 passed, test_cache_transceiver_precheck_config
55 passed.

Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: chenfeiz0326 <203214996+chenfeiz0326@users.noreply.github.com>