[
https://issues.apache.org/jira/browse/IMPALA-8565?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16857172#comment-16857172
]
Tim Armstrong commented on IMPALA-8565:
---------------------------------------
It looks like the query is pretty badly designed to reliably run slowly. Query
execution is super-fast, I think we're relying on the client fetching slowing
down the whole thing and applying backpressure. Maybe the one fragment finishes
fast because it races the other fragments and fills up the exchange buffer on
the coordinator. Or something. Anyway we should redesign the query
{noformat}
+---------------------+--------+----------+----------+---------+------------+----------+---------------+------------------------+
| Operator | #Hosts | Avg Time | Max Time | #Rows | Est. #Rows |
Peak Mem | Est. Peak Mem | Detail |
+---------------------+--------+----------+----------+---------+------------+----------+---------------+------------------------+
| F20:ROOT | 1 | 952.30ms | 952.30ms | | | 0
B | 0 B | |
| 20:EXCHANGE | 1 | 4.05ms | 4.05ms | 208.98K | 20.90K |
13.69 MB | 942.25 KB | UNPARTITIONED |
| F19:EXCHANGE SENDER | 3 | 19.74ms | 21.93ms | | |
1.48 KB | 0 B | |
| 00:UNION | 3 | 264.38us | 536.74us | 208.98K | 20.90K | 0
B | 0 B | |
| |--02:SCAN HDFS | 3 | 2.61ms | 3.93ms | 11.00K | 1.10K |
1.42 MB | 80.00 MB | functional.alltypesagg |
| |--03:SCAN HDFS | 3 | 2.69ms | 3.67ms | 11.00K | 1.10K |
1.25 MB | 80.00 MB | functional.alltypesagg |
| |--04:SCAN HDFS | 3 | 2.24ms | 3.49ms | 11.00K | 1.10K |
1.25 MB | 80.00 MB | functional.alltypesagg |
| |--05:SCAN HDFS | 3 | 2.83ms | 3.45ms | 11.00K | 1.10K |
1.15 MB | 80.00 MB | functional.alltypesagg |
| |--06:SCAN HDFS | 3 | 3.19ms | 5.03ms | 11.00K | 1.10K |
1.25 MB | 80.00 MB | functional.alltypesagg |
| |--07:SCAN HDFS | 3 | 2.39ms | 2.69ms | 11.00K | 1.10K |
1.59 MB | 80.00 MB | functional.alltypesagg |
| |--08:SCAN HDFS | 3 | 2.29ms | 3.17ms | 11.00K | 1.10K |
1.42 MB | 80.00 MB | functional.alltypesagg |
| |--09:SCAN HDFS | 3 | 3.26ms | 6.18ms | 11.00K | 1.10K |
1.30 MB | 80.00 MB | functional.alltypesagg |
| |--10:SCAN HDFS | 3 | 1.96ms | 2.47ms | 11.00K | 1.10K |
1.25 MB | 80.00 MB | functional.alltypesagg |
| |--11:SCAN HDFS | 3 | 2.59ms | 4.31ms | 11.00K | 1.10K |
1.59 MB | 80.00 MB | functional.alltypesagg |
| |--12:SCAN HDFS | 3 | 1.84ms | 2.01ms | 11.00K | 1.10K |
1.59 MB | 80.00 MB | functional.alltypesagg |
| |--13:SCAN HDFS | 3 | 1.73ms | 2.12ms | 11.00K | 1.10K |
1.61 MB | 80.00 MB | functional.alltypesagg |
| |--14:SCAN HDFS | 3 | 1.75ms | 1.86ms | 11.00K | 1.10K |
1.42 MB | 80.00 MB | functional.alltypesagg |
| |--15:SCAN HDFS | 3 | 1.78ms | 1.94ms | 11.00K | 1.10K |
1.42 MB | 80.00 MB | functional.alltypesagg |
| |--16:SCAN HDFS | 3 | 1.64ms | 1.75ms | 11.00K | 1.10K |
1.59 MB | 80.00 MB | functional.alltypesagg |
| |--17:SCAN HDFS | 3 | 1.44ms | 1.50ms | 11.00K | 1.10K |
1.59 MB | 80.00 MB | functional.alltypesagg |
| |--18:SCAN HDFS | 3 | 1.63ms | 1.77ms | 11.00K | 1.10K |
1.59 MB | 80.00 MB | functional.alltypesagg |
| |--19:SCAN HDFS | 3 | 1.53ms | 1.62ms | 11.00K | 1.10K |
1.59 MB | 80.00 MB | functional.alltypesagg |
| 01:SCAN HDFS | 3 | 2.98ms | 3.19ms | 11.00K | 1.10K |
1.25 MB | 80.00 MB | functional.alltypesagg |
+---------------------+--------+----------+----------+---------+------------+----------+---------------+------------------------+
{noformat}
> Test failure in TestAdmissionControllerStress.test_mem_limit
> ------------------------------------------------------------
>
> Key: IMPALA-8565
> URL: https://issues.apache.org/jira/browse/IMPALA-8565
> Project: IMPALA
> Issue Type: Bug
> Components: Infrastructure
> Reporter: David Rorke
> Assignee: Tim Armstrong
> Priority: Critical
> Labels: flaky
> Attachments:
> impalad.impala-ec2-centos74-m5-4xlarge-ondemand-04c0.vpc.cloudera.com.jenkins.log.INFO.20190518-000920.27854,
>
> impalad.impala-ec2-centos74-m5-4xlarge-ondemand-04c0.vpc.cloudera.com.jenkins.log.INFO.20190518-000920.27857,
>
> impalad.impala-ec2-centos74-m5-4xlarge-ondemand-04c0.vpc.cloudera.com.jenkins.log.INFO.20190518-000920.27863,
> test1.txt, test2.txt
>
>
> We had a recent test failure in test_mem_limit:
> {noformat}
> custom_cluster/test_admission_controller.py:1504: in test_mem_limit
> {'request_pool': self.pool_name, 'mem_limit': query_mem_limit})
> custom_cluster/test_admission_controller.py:1365: in run_admission_test
> assert metric_deltas['dequeued'] == 0,\
> E AssertionError: Queued queries should not run until others are made to
> finish
> E assert 3 == 0
> {noformat}
> {noformat}
> assert 7 >= 10 + where 10 = min((30 - 5), 10)
> Stacktrace
> custom_cluster/test_admission_controller.py:1504: in test_mem_limit
> {'request_pool': self.pool_name, 'mem_limit': query_mem_limit})
> custom_cluster/test_admission_controller.py:1384: in run_admission_test
> assert queries_page_num_queued >=\
> E assert 7 >= 10
> E + where 10 = min((30 - 5), 10)
> {noformat}
> It looks like this was a release build.
> [^test1.txt] was master
> [^test2.txt] was 2.12 with a bunch of cloudera patches
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscr...@impala.apache.org
For additional commands, e-mail: issues-all-h...@impala.apache.org
