Todd Lipcon has posted comments on this change. Change subject: Add more trace counters and timings ......................................................................
Patch Set 1: (1 comment) http://gerrit.cloudera.org:8080/#/c/7819/1/src/kudu/util/env_posix.cc File src/kudu/util/env_posix.cc: Line 764: virtual Status Read(uint64_t offset, Slice* result) const OVERRIDE { > Yeah but you don't like having traces at multiple levels? What if we have h I think my fear is that it becomes hard to know if the traces are additive or not. eg if it says "rwfile_read_time = 1sec" and "lbm_read_time = 1sec" should you expect those two add up together to a total of 2sec in the overall request latency? I don't have a particularly good general solution to this problem, though, aside from actually capturing these things on some kind of stack, but then performance of capturing/tracking becomes a concern. Perhaps we could maintain some kind of context "call stack" of limited/fixed depth so that these things end up netting out as counters like "cfile_write.lbm_write.rwfile_write_us"? -- To view, visit http://gerrit.cloudera.org:8080/7819 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-MessageType: comment Gerrit-Change-Id: Ie99bc9e06be682e1595745f154b7dded3903bd6e Gerrit-PatchSet: 1 Gerrit-Project: kudu Gerrit-Branch: master Gerrit-Owner: Todd Lipcon <[email protected]> Gerrit-Reviewer: Adar Dembo <[email protected]> Gerrit-Reviewer: Kudu Jenkins Gerrit-Reviewer: Tidy Bot Gerrit-Reviewer: Todd Lipcon <[email protected]> Gerrit-HasComments: Yes
