Alexey Serbin has posted comments on this change. ( http://gerrit.cloudera.org:8080/14897 )
Change subject: [master] KUDU-3016 flag for chunking tablet report updates ...................................................................... Patch Set 9: (3 comments) http://gerrit.cloudera.org:8080/#/c/14897/6/src/kudu/master/catalog_manager.cc File src/kudu/master/catalog_manager.cc: http://gerrit.cloudera.org:8080/#/c/14897/6/src/kudu/master/catalog_manager.cc@4305 PS6, Line 4305: : const SysCatalogTable::WriteMode write_mode = > Right; I was asking from a performance standpoint though: if chunk N fails It seems that according to this code https://github.com/apache/kudu/blob/0d7ce6906d42a17a7cfabc958e672ddc39e9ea7b/src/kudu/master/sys_catalog.cc#L830-L835 tablets which were updated by the visitor after first chunk had come through are not generating any updates on already updated rows upon processing the same tablet report again. http://gerrit.cloudera.org:8080/#/c/14897/9/src/kudu/master/catalog_manager.cc File src/kudu/master/catalog_manager.cc: http://gerrit.cloudera.org:8080/#/c/14897/9/src/kudu/master/catalog_manager.cc@282 PS9, Line 282: DEFINE_bool(catalog_manager_enable_chunked_tablet_reports, false, > Why shouldn't this default to true from day one? During offline discussion with the team I heard a concern w.r.t. the code paths when a chunked update fails in the middle. Also, simply rejecting whole tablet report seems to be in-line with rejecting DDL requests under similar conditions. However, after giving it another thought, I think that 100% rejecting a whole tablet report is even worse because there is no way the information in the system tablet can become consistent without updating the rpc_max_message_size, and it's hard to see the rejections are happening unless actively monitoring the newly introduced 'sys_catalog_oversized_write_requests' metric. Let's make it enabled by default. Thanks! http://gerrit.cloudera.org:8080/#/c/14897/9/src/kudu/master/catalog_manager.cc@4307 PS9, Line 4307: PREDICT_FALSE(FLAGS_catalog_manager_enable_chunked_tablet_reports) > PREDICT_FALSE seems like it'd become stale if/when the gflag's default valu Done -- To view, visit http://gerrit.cloudera.org:8080/14897 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-Project: kudu Gerrit-Branch: master Gerrit-MessageType: comment Gerrit-Change-Id: I83e8ca4bc8db7cab8fee6b4a40f48adc8752e7c5 Gerrit-Change-Number: 14897 Gerrit-PatchSet: 9 Gerrit-Owner: Alexey Serbin <aser...@cloudera.com> Gerrit-Reviewer: Adar Dembo <a...@cloudera.com> Gerrit-Reviewer: Alexey Serbin <aser...@cloudera.com> Gerrit-Reviewer: Andrew Wong <aw...@cloudera.com> Gerrit-Reviewer: Kudu Jenkins (120) Gerrit-Reviewer: Tidy Bot (241) Gerrit-Comment-Date: Tue, 14 Jan 2020 06:43:38 +0000 Gerrit-HasComments: Yes