Fix non-monotonic ProfileEvents increments

[azat]

NetworkReceiveBytes and AIOWriteBytes could be incremented with a negative ssize_t (-1 on socket timeout, -errno on AIO failure), wrapping the unsigned counter

MemoryOvercommitWaitTimeMicroseconds measured a duration with system_clock, which can step backwards. Guard the amounts and use steady_clock.

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

Changelog entry (a user-readable short description of the changes that goes into CHANGELOG.md):

Fix non-monotonic ProfileEvents increments

Version info

• Merged into: 26.7.1.481 (included in 26.7 and later)

[clickhouse-gh]

Workflow [PR], commit [5e255c2]

Summary: ❌

• Performance Comparison: Performance dashboard

job_name	test_name	status	info	comment
AST fuzzer (arm_asan_ubsan)		FAIL
	Logical error: Unexpected return type from A. Expected B. Got C (STID: 3262-40aa)	FAIL	cidb
Finish Workflow		FAIL
	python3 ./ci/jobs/scripts/workflow_hooks/new_tests_check.py	FAIL

---

AI Review

Summary

This PR fixes three sources of non-monotonic ProfileEvents by ignoring negative byte counts in the socket and AIO paths and by switching overcommit wait timing to steady_clock. The implementation looks correct, but the Bug Fix claim is still under-evidenced: there is no focused repro for the failure paths being fixed, so the wraparound could be reintroduced later without any test going red.

Tests

• ⚠️ Add at least one targeted regression that drives a negative ssize_t or timeout path and asserts the affected ProfileEvents counter stays monotonic. The easiest candidate is the ReadBufferFromPocoSocket timeout path, which already has nearby integration coverage in tests/integration/test_secure_socket/test.py; a failing-before/passing-after check there would give this fix concrete evidence.

Final Verdict

• Status: ⚠️ Request changes
• Minimum required action: add a focused regression for one of the fixed failure paths, preferably the socket-timeout path, or explain why the bug cannot be exercised in a testable way.

[clickhouse-gh]

LLVM Coverage Report

Metric	Baseline	Current	Δ
Lines	85.50%	85.50%	+0.00%
Functions	92.70%	92.70%	+0.00%
Branches	77.70%	77.70%	+0.00%

Changed lines: Changed C/C++ lines covered: 20/21 (95.24%) · Uncovered code

Full report · Diff report

[azat]

AST fuzzer (arm_asan_ubsan)

• Will take a look - https://pastila.nl/?00d9e66d/cbd99fbe0e3418efc8ddad4cdc9db3d0#TL71nh7n2q/epzKjSLnCGA==GCM
• Fix exception in correlated subquery with group_by_use_nulls + ROLLUP/CUBE #100365