Re: hive precommit tests on bigtop jenkins

2014-02-17 Thread Brock Noland
I agree that would be ideal. The challenge with Hive is that it must be distributed which adds to the complexity significantly. I'd love to see that added to ptest if someone has time. On Sun, Feb 16, 2014 at 11:59 PM, Thejas Nair wrote: > In theory, if there are 10 pending patches, we would comb

Re: hive precommit tests on bigtop jenkins

2014-02-16 Thread Thejas Nair
In theory, if there are 10 pending patches, we would combine the 10 into one patch (or find a group that applies together) and then run the tests once. Assuming the number of failures are a fraction of these tests, the failed tests can be then run with each of these patches one at a time. If the nu

Re: hive precommit tests on bigtop jenkins

2014-02-16 Thread Brock Noland
On Sun, Feb 16, 2014 at 11:11 PM, Navis류승우 wrote: > bq. even if a JIRA is in the queue twice it will only be tested once. > Good to know! > > bq. removing order-by clauses just for conforming purpose (my comment) > > I've tested it in https://issues.apache.org/jira/browse/HIVE-6438, making > 556 s

Re: hive precommit tests on bigtop jenkins

2014-02-16 Thread Navis류승우
bq. even if a JIRA is in the queue twice it will only be tested once. Good to know! bq. removing order-by clauses just for conforming purpose (my comment) I've tested it in https://issues.apache.org/jira/browse/HIVE-6438, making 556 sec -> 418 sec for join_filters.q. Would it be worthwhile to rew

Re: hive precommit tests on bigtop jenkins

2014-02-13 Thread Brock Noland
Hi, The pre-commit tests: 1) only test the latest attachment 2) post the attachment id to the JIRA 3) Verify the attachment id has not been tested before running This means that even if a JIRA is in the queue twice it will only be tested once. Below are relevant portions of the script: curl -s

Re: hive precommit tests on bigtop jenkins

2014-02-13 Thread Navis류승우
Recently, precommit test takes more than 1 day (including queue time). Deduping work queue (currently, HIVE-6403 and HIVE-6418 is queued twice) can make this better. Rewriting some test queries simpler (I'm thinking of removing order-by clauses just for conforming purpose). Any other ideas? 2014

Re: hive precommit tests on bigtop jenkins

2014-02-13 Thread Thejas Nair
I see a new job now running there. Maybe there is nothing wrong with the infra and builds actually finished (except for the 3 aborted ones). Can't complain about a shorter queue ! :) On Thu, Feb 13, 2014 at 1:30 PM, Thejas Nair wrote: > Is the jenkins infra used for hive precommit tests under