Reduce memory usage of bulkloader #4529
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
… key into a batch. That causes processing paused for a long time and the memory usage grows rapaidly. This commit fix it.
golangcibot
reviewed
Jan 9, 2020
dgraph/cmd/bulk/reduce.go
Outdated
| @@ -214,6 +239,8 @@ func (r *reducer) encodeAndWrite( | |||
| listSize = 0 | |||
| } | |||
| } | |||
| kvb.finish = true | |||
| listSize += r.toList(nil, list, kvb) | |||
There was a problem hiding this comment.
ineffectual assignment to listSize (from ineffassign)
dgraph/cmd/bulk/reduce.go
Outdated
| @@ -60,10 +67,18 @@ func (r *reducer) run() error { | |||
| go func(shardId int, db *badger.DB) { | |||
| defer thr.Done(nil) | |||
|
|
|||
| //Use a chan as mapEntries pool for mapIterator to read mapoutput. Fixed size is ok because mapEntries will | |||
xiangzhao632
changed the title
golangcibot
reviewed
Jan 11, 2020
| @@ -159,8 +166,7 @@ func newMapIterator(filename string) *mapIterator { | |||
| return &mapIterator{fd: fd, reader: bufio.NewReaderSize(gzReader, 16<<10)} | |||
| } | |||
|
|
|||
| func (r *reducer) encodeAndWrite( | |||
| writer *badger.StreamWriter, entryCh chan []*pb.MapEntry, closer *y.Closer) { | |||
| func (r *reducer) encodeAndWrite(writer *badger.StreamWriter, entryCh chan []*pb.MapEntry, closer *y.Closer) { | |||
|
|
1 similar comment
|
|
|
I have opened another pr for this. I really need to study how to use git and github. Close. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I used bulkloader to initial my cluster, but It was killed due to OOM. The dataset only has one predicate: dgraph.type, so I set reducer to 1, but the reduce stage still occupied 100G memory.
I located the problem at the reduce.go source file: all mapEntries with the same key must in one batch to produce postinglist. And in my dataset, postinglists can be very large. That leads to a long time to produce a batch and a very large batch size. So I make a little change to improve bulkloader.
The idea is that use a struct named kvBuilder to store inbuilding kv, So a batch doesn't have to include all all mapEntries with the same key, therefore saves memory from holding massive duplicate keys. This small change brings two benefits: first, process won't be block by a large batch, therefore speeds up bulkloader; second, memory usage will be reduced.
By this method, memory usage of the reduce stage cut down to 7G, and achieved 1.33x speedup.
This change is