Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Duplicate
-
4.5.1
-
None
-
None
-
Ubuntu 14.04 / 64-bit
-
Untriaged
-
Unknown
Description
I do have my own test suite (that includes 69 tests) for application and typical cycle of 1000-cyclic-runs of this suite to evaluate stability. With src-based 4.5.1-2844 (and JavaSDK 2.3.5) i see my tests die sporadically on arrayAppend results check (check is a 10 x sequence of "sync-get-doc / check-content / sleep 1s", so if after 10 seconds there is no needed content, it dies). Check shows, that there is a missing value (always one, but it can be a different value in a different documents within different tests in suite).
More:
- JavaSDK 2.3.5 and src-based build 4.5.1-2844: guaranteed fail on cyclic iteration 0..40
- JavaSDK 2.3.5 and src-based build 4.6.0-3544 there is NO problem for 1000 iterations
- I cannot reproduced this error with simple standalone arrayAppend-test
So, this is more like "please, keep this in mind", then "reproducable bug", because i'm not sure if this problem is gone for 4.6.0 or it just "hides behind some changes" in 4.6.0
Attachments
Issue Links
- duplicates
-
-
- Closed
-
Activity
Dave Finlay, you are right, https://issues.couchbase.com/browse/MB-21597 looks exact like my problem (and yes, appends are concurent). So, most likely my issue is a dup of https://issues.couchbase.com/browse/MB-21597.
Small remark: i could not reproduce it with 4.5.1-standalone test (not for 4.6; 4.6 has no problems for 1000-runs of my test suite, so, i did not try to make a standalone test for 4.6), but that's more like test related problem (remote host for standalone-test-run vs local lan for my suite run + specific problem with JavaSDK).
Close this please.
Georgy - I think you may have run into
MB-21597, a subdoc concurrency bug. This is fixed in 4.6 and is only visible under concurrent update across different clients (which may be why you are not seeing it against 4.6 and in your standalone test.) Does your test suite test array_append under concurrent update?