Lars Volker has posted comments on this change. Change subject: IMPALA-4494: Fix crash in SimpleScheduler ......................................................................
Patch Set 1: Thank you for your comments. I modified the change to always schedule at the coordinator if exec_at_coord is true. However, I think that scheduling at the coordinator if !exec_at_coord and no other backends are registered is not what we would want, since that way a large query can be completely scheduled at the coordinator, just because it has not received the initial update from the statestored. It is also the behavior of Impala 2.6 to abort scheduling in this case. There is still a small time window during daemon startup in which the statestore could have sent the initial update, to which the coordinator will reply by registering itself with the statestore. If we hit that window before the statestore sends back the local backend's registration, then the coordinator will not schedule scan ranges on itself. Should we change that, too? For example, we could add the local backend to the list of backends if it is non-empty but does not contain the local scheduler. -- To view, visit http://gerrit.cloudera.org:8080/5127 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-MessageType: comment Gerrit-Change-Id: I6e1196a2fa47e5954c4a190aa326c135d039a77f Gerrit-PatchSet: 1 Gerrit-Project: Impala-ASF Gerrit-Branch: master Gerrit-Owner: Lars Volker <l...@cloudera.com> Gerrit-Reviewer: Henry Robinson <he...@cloudera.com> Gerrit-Reviewer: Lars Volker <l...@cloudera.com> Gerrit-Reviewer: Tim Armstrong <tarmstr...@cloudera.com> Gerrit-HasComments: No