Tim Armstrong has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/16549 )

Change subject: IMPALA-6671: [WIP] Skip locked tables from topic updates
......................................................................


Patch Set 1:

(3 comments)

I think if this approach works out it would meaningfully improve metadata for 
production workloads - seems like it should reduce the number of hiccups that 
non-long-running-ddl queries experience, although obviously with the approach 
they can eventually get stuck behind a DDL.

We'd expect that potentially, this could have some impact on tables where there 
are multiple long-running DDLs, right? E.g. that table might skip topic updates 
and delay queries waiting for that particular table. I think that seems fine, 
but just wanted to be sure.

http://gerrit.cloudera.org:8080/#/c/16549/1//COMMIT_MSG
Commit Message:

http://gerrit.cloudera.org:8080/#/c/16549/1//COMMIT_MSG@9
PS1, Line 9: This change adds a mechanism for topic-update thread
question, that maybe will be answered as I read the patch - if an update is 
skipped, does the topic update collection thread get woken up when the lock is 
released?


http://gerrit.cloudera.org:8080/#/c/16549/1//COMMIT_MSG@23
PS1, Line 23: This solution unblocks topic-updates for a while until a
Maybe this is the best we can do at the moment, but it does seem to have 
similar problems to the IMPALA-5058 solution where it does help with latency 
when it's blocked for shorter periods of time but still gets blocked when a 
lock is held for a long time.


http://gerrit.cloudera.org:8080/#/c/16549/1/fe/src/main/java/org/apache/impala/catalog/CatalogServiceCatalog.java
File fe/src/main/java/org/apache/impala/catalog/CatalogServiceCatalog.java:

http://gerrit.cloudera.org:8080/#/c/16549/1/fe/src/main/java/org/apache/impala/catalog/CatalogServiceCatalog.java@1392
PS1, Line 1392:     if (tblVersion > ctx.toVersion &&
I've seen some cases in production workloads where the topic update thread gets 
stuck acquiring the table lock, but then has to skip the update anywhere here. 
Is there any way we can move this check outside of the lock acquisition so that 
it can exit early in that case?



--
To view, visit http://gerrit.cloudera.org:8080/16549
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: Ic657b96edbcdc94c6b906e7ca59291f4e4715655
Gerrit-Change-Number: 16549
Gerrit-PatchSet: 1
Gerrit-Owner: Vihang Karajgaonkar <vih...@cloudera.com>
Gerrit-Reviewer: Impala Public Jenkins <impala-public-jenk...@cloudera.com>
Gerrit-Reviewer: Shant Hovsepian <sh...@cloudera.com>
Gerrit-Reviewer: Tim Armstrong <tarmstr...@cloudera.com>
Gerrit-Reviewer: Vihang Karajgaonkar <vih...@cloudera.com>
Gerrit-Comment-Date: Tue, 06 Oct 2020 19:50:09 +0000
Gerrit-HasComments: Yes

Reply via email to