Dan Burkert has posted comments on this change. Change subject: KUDU-2125: Tablet copy client does not retry on failures ......................................................................
Patch Set 10: (3 comments) http://gerrit.cloudera.org:8080/#/c/8016/10/src/kudu/integration-tests/cluster_itest_util.cc File src/kudu/integration-tests/cluster_itest_util.cc: PS10, Line 858: != > Neat: I didn't know you could check for equality between a thing and an opt Yah, it auto derefs it. Very dangerous if you don't do a check up front, though. I don't think it throws an exception or anything. http://gerrit.cloudera.org:8080/#/c/8016/10/src/kudu/tserver/tablet_copy_client.cc File src/kudu/tserver/tablet_copy_client.cc: PS10, Line 702: >= > why >= and not == ? I thought about it, and mildly preferred >= so that the client doesn't continue looping indefinitely if the server sends a block with a too-long length. I'm not sure what would happen in that scenario, but infinite loop doesn't seem good. Line 747: // Polynomial backoff with 50% jitter. > Have you used this before? I plotted y=10x^2 and it seems pretty good -- we I briefly tried to see if we had an abstraction for exponential backoff with jitter calculation, and it seems we don't. This was just gut feeling. I also graphed it initially, for anyone else following along: https://www.wolframcloud.com/objects/b532668a-64c0-45d4-8fad-dfce1718cda6 -- To view, visit http://gerrit.cloudera.org:8080/8016 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-MessageType: comment Gerrit-Change-Id: I7c8454fc600a841bd15306a2b3b06ddf53130be6 Gerrit-PatchSet: 10 Gerrit-Project: kudu Gerrit-Branch: master Gerrit-Owner: Dan Burkert <danburk...@apache.org> Gerrit-Reviewer: Adar Dembo <a...@cloudera.com> Gerrit-Reviewer: Alexey Serbin <aser...@cloudera.com> Gerrit-Reviewer: Dan Burkert <danburk...@apache.org> Gerrit-Reviewer: Kudu Jenkins Gerrit-Reviewer: Mike Percy <mpe...@apache.org> Gerrit-Reviewer: Todd Lipcon <t...@apache.org> Gerrit-HasComments: Yes