test/access_loggers/stats: replace tcmalloc consumedBytes assertions with sizeof structural bounds

[Copilot]

GaugeKeyTest.ExactMemoryFootprint and StatsAccessLoggerTest.AccessLogStateMemoryFootprint used Memory::TestUtil::MemoryTest::consumedBytes() with hand-tuned tcmalloc fudge factors ("384 gives a generous 1.5x buffer for allocator page alignment", "We allow 4096 bytes for allocator caches"). This is the same anti-pattern removed for StringMatcher in #44701 — it produces CI flakes and needs re-tuning on every tcmalloc/abseil/protobuf bump.

Changes

• GaugeKeyTest::ExactMemoryFootprint → GaugeKeyTest::SizeIsBounded: drop all three MemoryTest/consumedBytes() blocks; replace with a single portable structural bound:

  EXPECT_THAT(sizeof(GaugeKey), MemNotMoreThan(8 * sizeof(void*)));

• StatsAccessLoggerTest::AccessLogStateMemoryFootprint → StatsAccessLoggerTest::AccessLogStateSizeIsBounded: remove the 10 K-item insertion/removal loop and all consumedBytes() assertions (including the two that bypassed EXPECT_MEMORY_LE with raw EXPECT_LE); replace with:

  EXPECT_THAT(sizeof(AccessLogState), MemNotMoreThan(16 * sizeof(void*)));
  EXPECT_THAT(sizeof(GaugeKey),       MemNotMoreThan(8 * sizeof(void*)));

  Comments document the InflightGauge layout (private struct) and explain that leak-after-destruction is already covered by ASAN/MSAN/LSAN in CI.

• Add MATCHER_P(MemNotMoreThan, sz, ...) (identical to test: replace flaky StringMatcher.Memory with deterministic sizeof bounds #44701): failure message prompts the reviewer to think carefully before raising the bound.

• Remove #include "test/common/memory/memory_test_utility.h" — no remaining callers.

All bounds are expressed in sizeof(void*) units for portability across libc++/libstdc++ and 32/64-bit builds.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• dl.google.com
  • Triggering command: /build/bazel_root/install/fb2a7f6d344d2f4e335882534df59296/embedded_tools/jdk/bin/java bazel(envoy) --add-opens=java.base/java.lang=ALL-UNNAMED -Xverify:none -Djava.util.logging.config.file=/build/bazel_root/base/javalog.properties -Dcom.google.devtools.build.lib.util.LogHandlerQuerier.class=com.google.devtools.build.lib.util.SimpleLogHandler$HandlerQuerier -XX:-MaxFDLimit -Djava.library.path=/build/bazel_root/install/fb2a7f6d344d2f4e335882534df59296/embedded_tools/jdk/lib:/build/bazel_root/install/fb2a7f6d344d2f4e335882534df59296/embedded_tools/jdk/lib/server:/build/bazel_root/install/fb2a7f6d344d2f4e335882534df59296/ -Dfile.encoding=ISO-8859-1 -Duser.country= -Duser.language= -Duser.variant= -Xmx3g -DBAZEL_TRACK_SOURCE_DIRECTORIES=1 -Djavax.net.ssl.trustStore=/tmp/custom-cacerts -Djavax.net.ssl.trustStorePassword=changeit -jar /build/bazel_root/install/fb2a7f6d344d2f4e335882534df59296/A-server.jar --max_idle_secs=10800 --noshutdown_on_low_sys_mem --connect_timeout_secs=30 (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

Original prompt

Background

PR #44701 replaced TEST_F(StringMatcher, Memory) with StringMatcher.SizeIsBounded, removing tcmalloc page-level consumedBytes() assertions with hand-tuned ceilings (which produced ~5/1000 CI flakes and had to be re-tuned on every tcmalloc/abseil/protobuf bump — see #37782, #43467, #44528).

The same anti-pattern exists in test/extensions/access_loggers/stats/stats_test.cc, in two places:

1. A small EXPECT_MEMORY_LE(memory_test.consumedBytes(), 64); block (just before AccessLogStateMemoryFootprint).
2. TEST_F(StatsAccessLoggerTest, AccessLogStateMemoryFootprint) — which mixes a deterministic EXPECT_LE(sizeof(AccessLogState), 128) static check with multiple consumedBytes()-based assertions whose comments explicitly call out allocator fudge factors:
   • "384 gives a generous 1.5x buffer for allocator page alignment."
   • "We allow 4096 bytes for allocator caches."
   • "approximately 1.6 Megabytes for 10,000 items ... threshold of 2 Megabytes here to account for this capacity and allocator page alignment."

It also bypasses EXPECT_MEMORY_LE in two spots with raw EXPECT_LE(static_cast<int64_t>(memory_test.consumedBytes()), …) and EXPECT_LE(memory_test.consumedBytes(), 4096), which skips the canonical-platform guard that EXPECT_MEMORY_LE provides — so on non-canonical platforms those lines are effectively asserting tcmalloc page math directly, which is exactly the flake source we want to remove.

The test is young (added Feb 2026 in #42226) so it hasn't been re-tuned yet, but it will start needing the same toolchain-bump tuning that #44701 retired.

Task

Apply the same approach used in #44701 to test/extensions/access_loggers/stats/stats_test.cc:

• Replace the Memory::TestUtil::MemoryTest / consumedBytes() based assertions with deterministic sizeof-based structural bounds, expressed in sizeof(std::string) / sizeof(void*) / sizeof(some_type) for libc++ / libstdc++ / 32-bit / 64-bit portability.
• Use a MATCHER_P(MemNotMoreThan, sz, ...) (or equivalent) similar to the one introduced in PR test: replace flaky StringMatcher.Memory with deterministic sizeof bounds #44701 so the failure message tells the maintainer to "think carefully before increasing this, and if you're sure, update the corresponding expectation."
• Preserve the intent of each existing assertion — i.e. the bound should still describe the structural footprint the test was trying to gate on (e.g. that AccessLogState and the GaugeKey/InflightGauge slot types stay small). The existing comments documenting why the bounds exist (slot composition, what the per-item overhead is, etc.) should be kept and re-cast in terms of sizeof of the relevant types instead of measured tcmalloc bytes.
• Where the existing test was checking "no leak after destruction" via consumedBytes(), drop the tcmalloc-based check entirely if it cannot be re-expressed deterministically. The destructor invariants the test was trying to assert are better covered by ASAN/MSAN/LSAN, which already run on this code in CI; preserving a flaky leak check here adds no real safety net. Add a brief comment explaining this.
• Keep the test function name(s) descriptive of what they actually now check (e.g. rename AccessLogStateMemoryFootprint to AccessLogStateSizeIsBounded or similar, mirroring the rename in test: replace flaky StringMatcher.Memory with deterministic sizeof bounds #44701).
• Remove #include "test/common/memory/memory_test_utility.h" if no other test in the file still needs it after the change.
• If gmock/gmock.h is not already included, add it (needed for EXPECT_THAT / MATCHER_P).

Reference

The PR to mirror: #44701 — see the diff to test/common/common/matchers_test.cc for the exact style, the MemNotMoreThan matcher, and the bound expressions to follow.

Out of scope

• Do not touch the legitimate regression-gating memory tests (test/integration/stats_integration_test.cc ClusterMemoryTest, test/common/stats/thread_local_store_test.cc MemoryWith*RealSymbolTable, test/common/stats/symbol_table_impl_test.cc SymbolTableTest.Memory). Those are intentional golden-value trackers and are correctly using EXPECT_MEMORY_EQ / EXPECT_MEMORY_LE against pinned values.
• Do not touch test/common/common/packed_struct_test.cc in this PR.

Validation

• bazel test //test/extensions/access_loggers/stats:stats_test should pass.
• Confirm no remaining uses of Memory::TestUtil::MemoryTest / consumedBytes() / EXPECT_MEMORY_LE / EXPECT_MEMORY_EQ in the modified file.
• Format/lint per repo conventions.

The following is the prior conversation context from the user's chat exploration (may be truncated):

User: we just landed this one - which i think is a great improvement

i think there is another similar (perhaps ~adjacent) test that does the same anti-pattern and should be fixed

Assistant: Looking at the PR you just landed, the anti-pattern was: **using Memory::TestUtil::MemoryTest::consumedBytes() (tcmalloc page-level accounting) with hand-tuned ceilings + alignment fu...

This pull request was created from Copilot chat.

[repokitteh-read-only]

As a reminder, PRs marked as draft will not be automatically assigned reviewers,
or be handled by maintainer-oncall triage.

Please mark your PR as ready when you want it to be reviewed!

🐱

Caused by: #44708 was opened by Copilot.

see: more, trace.