allthingssecurity commented on PR #30352:
URL: https://github.com/apache/camel/pull/30352#issuecomment-7002916759

   Thanks Claus, all items are addressed in four follow-up commits (2861cdf, 
9598441, 611ad50, 2ccc19d).
   
   **1. Validation gate.** I ran the steps instead of moving them, with a 
window at every point (`overlapSampling=1`) and the CI switch:
   
   - netty 162 tests, netty-http 383, vertx-http 124: no failures; 561 contexts 
opened 43,689 windows (41,383 / 1,985 / 321) and reported nothing.
   - camel-jms request/reply (`*RequestReply*`, `*InOut*`, `*ReplyTo*` on the 
embedded Artemis broker): 110 tests, no failures; 81 contexts opened 16,546 
windows and reported nothing.
   - No `USED_BY_TWO_THREADS` anywhere. The only sanitizer output is two R1 
"still pending" warnings from two vertx-http tests that stop their context with 
an exchange in flight on purpose.
   - TLA+ model of the windows, TLC with no error: 50,700, 326,448 and 
1,177,940 distinct states for the three configurations. Four mutants are each 
caught by an invariant and four witnesses are reachable.
   - JMH, short and indicative, per message for the 6 steps route: off 0.94 us, 
sampling 0 1.20 us, sampling 100 183 us, sampling 1 18.7 ms (split route: 4.1 
us, 4.8 us, 306 us, 31 ms). The cost of the windows is the pause, 12 points x 
about 1.5 ms / sampling.
   - One explicit, separate decision: the full-length JMH run and the benchmark 
in camel-performance-tests move to after merge, as the windows are off outside 
tests and the short run already shows the cost is the pause. It is written in 
the PR body and in the design's Validation section (2ccc19d), which now has all 
the results.
   
   **2. Dev features together.** `ChaosWithSanitizerWatchWindowsTest` (611ad50) 
runs with `@CamelSanitizer(overlapSampling = 1)` and the tracer, backlog tracer 
and backlog debugger on. Its tests cover a step failed by chaos, a delayed 
step, and the `duplicate` rule on a send that completes synchronously, one that 
completes asynchronously, and one on a pool thread. Each asserts that windows 
opened and nothing was reported. The two chaos sanitizer classes 10 times: 110 
of 110. The camel-test-main-junit6 sanitizer tests pass with the default 
sampling (4 of 4).
   
   **Smaller items**
   
   - `sanitizer.adoc` now says accesses between the points are not seen, with 
your three examples (9598441).
   - The rerun sentence now holds only for one thread: with several threads the 
shared counter falls on other points. Changed in `Overlaps`, `sanitizer.adoc`, 
the camel-main option and the design (9598441, 2ccc19d).
   - `@CamelSanitizer(overlapSampling = -5)` now fails the test class with a 
message, before the extension takes it over (2861cdf).
   - The `camel.sanitizer.checks` description names the watch windows under 
`threads`. Main docs, metadata and catalog copies are regenerated (9598441).
   - `WatchWindowsByDefault` no longer NPEs with 
`-Dcamel.sanitizer.overlapSampling=0` (2861cdf): 24 of 24 in 3 runs with that 
property, and 80 of 80 in 10 runs without it.
   
   camel-core sanitizer tests 168 of 168 and camel-main sanitizer tests 10 of 
10 after the changes.
   
   _Claude Code on behalf of allthingssecurity_
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to