[
https://issues.apache.org/jira/browse/SPARK-21425?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16088868#comment-16088868
]
Sean Owen commented on SPARK-21425:
-----------------------------------
I'd expect the implementations to be prepared for concurrent writes, but
they're not. The docs talk a little more narrowly about making sure the output
value is prepared to be read concurrently.
My guess is the assumption is that only one thread updates the accumulators,
and that is consistent with my mental model that each task (single-threaded)
has one accumulator. Your example doesn't show a reproduction in Spark, but in
direct concurrent usage of the class.
Still I wonder if there's a corner case here where one task can use multiple
CPUs, but that's not what this reproduces.
Do you observe any problem in normal Spark usage? It does look like this was on
purpose, at least at one time.
CC [~rxin]
> LongAccumulator, DoubleAccumulator not threadsafe
> -------------------------------------------------
>
> Key: SPARK-21425
> URL: https://issues.apache.org/jira/browse/SPARK-21425
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 2.2.0
> Reporter: Ryan Williams
>
> [AccumulatorV2
> docs|https://github.com/apache/spark/blob/v2.2.0/core/src/main/scala/org/apache/spark/util/AccumulatorV2.scala#L42-L43]
> acknowledge that accumulators must be concurrent-read-safe, but afaict they
> must also be concurrent-write-safe.
> The same docs imply that {{Int}} and {{Long}} meet either/both of these
> criteria, when afaict they do not.
> Relatedly, the provided
> [LongAccumulator|https://github.com/apache/spark/blob/v2.2.0/core/src/main/scala/org/apache/spark/util/AccumulatorV2.scala#L291]
> and
> [DoubleAccumulator|https://github.com/apache/spark/blob/v2.2.0/core/src/main/scala/org/apache/spark/util/AccumulatorV2.scala#L370]
> are not thread-safe, and should be expected to behave undefinedly when
> multiple concurrent tasks on the same executor write to them.
> [Here is a repro repo|https://github.com/ryan-williams/spark-bugs/tree/accum]
> with some simple applications that demonstrate incorrect results from
> {{LongAccumulator}}'s.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
