drill-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ben-Zvi <...@git.apache.org>
Subject [GitHub] drill pull request #822: DRILL-5457: Spill implementation for Hash Aggregate
Date Fri, 02 Jun 2017 21:17:50 GMT
Github user Ben-Zvi commented on a diff in the pull request:

    https://github.com/apache/drill/pull/822#discussion_r119954158
  
    --- Diff: exec/java-exec/src/main/java/org/apache/drill/exec/physical/impl/aggregate/HashAggTemplate.java
---
    @@ -400,114 +782,411 @@ public IterOutcome getOutcome() {
     
       @Override
       public int getOutputCount() {
    -    // return outputCount;
         return lastBatchOutputCount;
       }
     
       @Override
       public void cleanup() {
    -    if (htable != null) {
    -      htable.clear();
    -      htable = null;
    -    }
    +      if ( schema == null ) { return; } // not set up; nothing to clean
    +      for ( int i = 0; i < numPartitions; i++) {
    +          if (htables[i] != null) {
    +              htables[i].clear();
    +              htables[i] = null;
    +          }
    +          if ( batchHolders[i] != null) {
    +              for (BatchHolder bh : batchHolders[i]) {
    +                    bh.clear();
    +              }
    +              batchHolders[i].clear();
    +              batchHolders[i] = null;
    +          }
    +
    +          // delete any (still active) output spill file
    +          if ( outputStream[i] != null && spillFiles[i] != null) {
    +            try {
    +              spillSet.delete(spillFiles[i]);
    --- End diff --
    
    Concurrent open files: While spilling, there is one per each (non-pristine) spilling partition
(yes, can be as high as 16, or even 32). Afterwards, they are all closed; then for reading,
each one gets opened; and though we process one partition at a time, closing of all is postponed
to the end, as the processing code is unaware that the "incoming" actually comes from a spill
file. 
        About the limits: Seems that current defaults (e.g. 64K open files per process) can
serve us well for the foreseeable future. Intel just announced the i9, where the top of the
line CPU has 18 cores. Hence 1000s of concurrent active same-process threads are not feasible
anytime soon (think about context switching). 


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

Mime
View raw message