Michael Ho has posted comments on this change. Change subject: IMPALA-2550: Switch to per-query exec rpc ......................................................................
Patch Set 4: (12 comments) http://gerrit.cloudera.org:8080/#/c/6535/4/be/src/runtime/coordinator-backend-state.cc File be/src/runtime/coordinator-backend-state.cc: PS4, Line 434: if (!status.ok()) return false; Does this warrant a log message ? It's not entirely clear what's the status of the remote fragment instances at this point, right ? The remote fragment instances could still be running. The log message may make debugging in the field slightly easier. http://gerrit.cloudera.org:8080/#/c/6535/4/be/src/runtime/coordinator.cc File be/src/runtime/coordinator.cc: PS4, Line 187: FinishBackendStartup Would a better name be VerifyBackendStartup(); ? http://gerrit.cloudera.org:8080/#/c/6535/4/be/src/runtime/fragment-instance-state.cc File be/src/runtime/fragment-instance-state.cc: PS4, Line 249: report_thread_.reset( : new Thread("plan-fragment-executor", "report-profile", : &FragmentInstanceState::ReportProfileThread, this)); Is one reporting thread per fragment instance a bit much ? Have you considered using one reporting thread per fragment ? PS4, Line 267: //RETURN_IF_ERROR(quer delete PS4, Line 298: RETURN_IF_ERROR(status); Why not RETURN_IF_ERROR(exec_tree_->GetNext(...))) at line 294 instead ? http://gerrit.cloudera.org:8080/#/c/6535/4/be/src/runtime/fragment-instance-state.h File be/src/runtime/fragment-instance-state.h: PS4, Line 83: void Cancel(); Please also document whether this is idempotent. PS4, Line 89: If the preparation phase encountered an error, GetOpenStatus() will : /// return that error without blocking. I don't see the opened_promise_ being updated when Prepare() fails in Exec(). Is that expected ? http://gerrit.cloudera.org:8080/#/c/6535/4/be/src/runtime/query-state.cc File be/src/runtime/query-state.cc: PS4, Line 194: stringstream s; Did you mean to use s somewhere ? It looks like dead code as it stands now. PS4, Line 246: 3 How is this determined ? Should we use #define or constexpr to represent it so it has a more meaningful name ? PS4, Line 250: rpc_status = Status(res.status); Is this needed given line 256 below which assigns res.status to result_status ? PS4, Line 317: Status status Not an error but there is a variable named status already in the outer scope. May want to have a more specific name. http://gerrit.cloudera.org:8080/#/c/6535/4/be/src/runtime/query-state.h File be/src/runtime/query-state.h: PS4, Line 172: AtomicInt32 is_cancelled_; > Is there any atomic read-modify-write operation on this flag ? Doesn't seem To answer my own question, there is an AtomicCompareAndSwap() operations on this flag to avoid calling cancellation on fragments multiple times. -- To view, visit http://gerrit.cloudera.org:8080/6535 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-MessageType: comment Gerrit-Change-Id: I20769e420711737b6b385c744cef4851cee3facd Gerrit-PatchSet: 4 Gerrit-Project: Impala-ASF Gerrit-Branch: master Gerrit-Owner: Marcel Kornacker <mar...@cloudera.com> Gerrit-Reviewer: Dan Hecht <dhe...@cloudera.com> Gerrit-Reviewer: Henry Robinson <he...@cloudera.com> Gerrit-Reviewer: Marcel Kornacker <mar...@cloudera.com> Gerrit-Reviewer: Michael Ho <k...@cloudera.com> Gerrit-HasComments: Yes
