[CI] InternalAutoDateHistogramTests::testReduceRandom fails reproducable for certain seed #39497
Comments
hendrikmuhs
added
:Analytics/Aggregations
|
Pinging @elastic/es-analytics-geo |
|
Also fails for ➜ elasticsearch git:(master) ./gradlew :server:unitTest -Dtests.seed=7DBCFCFC5EFB7BF8 -Dtests.class=org.elasticsearch.search.aggregations.bucket.histogram.InternalAutoDateHistogramTests -Dtests.method="testReduceRandom" -Dtests.security.manager=true -Dtests.locale=fil-PH -Dtests.timezone=Asia/Krasnoyarsk -Dcompiler.java=11 -Druntime.java=11
[junit4] JVM J0: 0.26 .. 1.76 = 1.51s FAILURE: Build failed with an exception.
|
|
This failed in a PR too #38382. This should be muted |
…ending resolution of elastic#39497
|
Found a case of this failing in I'm going to backport the mute to the |
|
Thanks for the mute, I'll take a look at this one. |
|
@polyfractal I've just been in this code, I'll have a look at this one |
|
I've been attempting to repro this issue locally with no luck, by un-muting
Neither of these cases, reported by @hendrikmuhs and @original-brownbear, repro for me after letting the above run for a long time. So, given that brute force reproducing has not worked, we must fall back to... logic and reasoning. Both these failures are in this kinda scenario: |
|
Closing, as I believe the fix in #44461 covers the cases highlighted in this. of course we will reopen this if my assertion is incorrect. |
|
The failure that occurred is of a different variety than earlier failures, the earlier failures were of the form "we expected one bucket but we got two", the latest failure is due to an incorrect innerInterval calculation: Definitely worth digging into, just wanted to mention that this latest error is a bit different nature than earlier errors, which were probably due to the bug we found. |
The test had errors around time units that have different length - think leap years or months that aren't 30 days. This fixes those errors. In the proces I've changed a bunch of things to debug the problem: * Replace `currentTimeMillis` with a random time. Now the test fails randomly! Wonderful. Much better than on random days of the month. * Generate buckets "closer together" to test random reduction. Without this we were super frequently getting stuck in the "year of century" rounding because *some* of the of the buckets we built were far apart. This generates a much greater variety of tests. * Implement `toString` on `RoundingInfo` so I can debug without going crazy. * Switch keys in the bucket assertions from epoch millis to `Instant`s so we can read the failures. Closes elastic#54540 Closes elastic#39497
The test had errors around time units that have different length - think leap years or months that aren't 30 days. This fixes those errors. In the proces I've changed a bunch of things to debug the problem: * Replace `currentTimeMillis` with a random time. Now the test fails randomly! Wonderful. Much better than on random days of the month. * Generate buckets "closer together" to test random reduction. Without this we were super frequently getting stuck in the "year of century" rounding because *some* of the of the buckets we built were far apart. This generates a much greater variety of tests. * Implement `toString` on `RoundingInfo` so I can debug without going crazy. * Switch keys in the bucket assertions from epoch millis to `Instant`s so we can read the failures. Closes #54540 Closes #39497
The test had errors around time units that have different length - think leap years or months that aren't 30 days. This fixes those errors. In the proces I've changed a bunch of things to debug the problem: * Replace `currentTimeMillis` with a random time. Now the test fails randomly! Wonderful. Much better than on random days of the month. * Generate buckets "closer together" to test random reduction. Without this we were super frequently getting stuck in the "year of century" rounding because *some* of the of the buckets we built were far apart. This generates a much greater variety of tests. * Implement `toString` on `RoundingInfo` so I can debug without going crazy. * Switch keys in the bucket assertions from epoch millis to `Instant`s so we can read the failures. Closes elastic#54540 Closes elastic#39497
The test had errors around time units that have different length - think leap years or months that aren't 30 days. This fixes those errors. In the proces I've changed a bunch of things to debug the problem: * Replace `currentTimeMillis` with a random time. Now the test fails randomly! Wonderful. Much better than on random days of the month. * Generate buckets "closer together" to test random reduction. Without this we were super frequently getting stuck in the "year of century" rounding because *some* of the of the buckets we built were far apart. This generates a much greater variety of tests. * Implement `toString` on `RoundingInfo` so I can debug without going crazy. * Switch keys in the bucket assertions from epoch millis to `Instant`s so we can read the failures. Closes elastic#54540 Closes elastic#39497
The test had errors around time units that have different length - think leap years or months that aren't 30 days. This fixes those errors. In the proces I've changed a bunch of things to debug the problem: * Replace `currentTimeMillis` with a random time. Now the test fails randomly! Wonderful. Much better than on random days of the month. * Generate buckets "closer together" to test random reduction. Without this we were super frequently getting stuck in the "year of century" rounding because *some* of the of the buckets we built were far apart. This generates a much greater variety of tests. * Implement `toString` on `RoundingInfo` so I can debug without going crazy. * Switch keys in the bucket assertions from epoch millis to `Instant`s so we can read the failures. Closes #54540 Closes #39497
The test had errors around time units that have different length - think leap years or months that aren't 30 days. This fixes those errors. In the proces I've changed a bunch of things to debug the problem: * Replace `currentTimeMillis` with a random time. Now the test fails randomly! Wonderful. Much better than on random days of the month. * Generate buckets "closer together" to test random reduction. Without this we were super frequently getting stuck in the "year of century" rounding because *some* of the of the buckets we built were far apart. This generates a much greater variety of tests. * Implement `toString` on `RoundingInfo` so I can debug without going crazy. * Switch keys in the bucket assertions from epoch millis to `Instant`s so we can read the failures. Closes #54540 Closes #39497
The following test failed in a PR-build and is reproducible for me on my local box:
fails with:
Tested only on master but likely applies to other versions.
The text was updated successfully, but these errors were encountered: