Henry Robinson has posted comments on this change. Change subject: IMPALA-3610: Account for memory used by filters in the coordinator ......................................................................
Patch Set 3: (3 comments) http://gerrit.cloudera.org:8080/#/c/3618/3/be/src/runtime/coordinator.cc File be/src/runtime/coordinator.cc: PS3, Line 2034: state->disabled = true; > Given the description of this variable, this should go inside DiscardFilter I don't think so: a filter may be discarded because it is complete, but that doesn't mean it was ever disabled. PS3, Line 2107: state->bloom_filter.reset(new BloomFilter(params.bloom_filter)); > Thinking about this more, why do we even need to do this for a broadcast jo That's a good point. We'd still need to copy the structure into rpc_params, but could get away without one alloc / copy. I'll see if that makes the code simpler. PS3, Line 2124: if (target.is_local) continue; > Is this a bug? What if it's a partitioned join filter and it's local to one I'm not quite following. If one fragment instance is local for a target, all fragment instances for that target will be local, since all those will be instances of the same fragment. So isn't it correct to skip publishing a filter to that target? Note that it's not possible to have local partitioned filters, because you can't publish a partition filter without aggregating all the other instances first. -- To view, visit http://gerrit.cloudera.org:8080/3618 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-MessageType: comment Gerrit-Change-Id: I3c52c8a1c2e79ef370c77bf264885fc859678d1b Gerrit-PatchSet: 3 Gerrit-Project: Impala Gerrit-Branch: cdh5-trunk Gerrit-Owner: Henry Robinson <[email protected]> Gerrit-Reviewer: Dan Hecht <[email protected]> Gerrit-Reviewer: Henry Robinson <[email protected]> Gerrit-Reviewer: Mostafa Mokhtar <[email protected]> Gerrit-Reviewer: Sailesh Mukil <[email protected]> Gerrit-HasComments: Yes
