You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@spark.apache.org by "wankunde (via GitHub)" <gi...@apache.org> on 2023/06/20 15:14:31 UTC
[GitHub] [spark] wankunde opened a new pull request, #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
wankunde opened a new pull request, #41676:
URL: https://github.com/apache/spark/pull/41676
<!--
Thanks for sending a pull request! Here are some tips for you:
1. If this is your first time, please read our contributor guidelines: https://spark.apache.org/contributing.html
2. Ensure you have added or run the appropriate tests for your PR: https://spark.apache.org/developer-tools.html
3. If the PR is unfinished, add '[WIP]' in your PR title, e.g., '[WIP][SPARK-XXXX] Your PR title ...'.
4. Be sure to keep the PR description updated to reflect all changes.
5. Please write your PR title to summarize what this PR proposes.
6. If possible, provide a concise example to reproduce the issue for a faster review.
7. If you want to add a new configuration, please read the guideline first for naming configurations in
'core/src/main/scala/org/apache/spark/internal/config/ConfigEntry.scala'.
8. If you want to add or modify an error type or message, please read the guideline first in
'core/src/main/resources/error/README.md'.
-->
### What changes were proposed in this pull request?
DAGScheduler will get the preferred locations for each RDD partition and try to allocate the task on the preferred locations.
We can remove the duplicate preferred locations to save memory.
For example. reduce 0 needs to fetch map0 output and map1 output in host-A, then the preferred locations can be Array("host-A").
### Why are the changes needed?
To save driver memory for large job, too many preferred locations may cause OOM.
```
2023-06-19T14:14:18.501-0700: [Full GC (Allocation Failure) 2023-06-19T14:14:18.501-0700: [Class Histogram (before full gc):
num #instances #bytes class name
----------------------------------------------
1: 2980686820 95381978240 scala.collection.immutable.$colon$colon
2: 2978345818 71480299632 org.apache.spark.scheduler.HostTaskLocation
3: 24388384 3963604560 [C
4: 24265128 776484096 java.lang.String
5: 368944 560810384 [Ljava.lang.Object;
6: 2927846 421609824 java.net.URI
7: 16239585 389750040 java.lang.Integer
8: 2540273 243866208 org.apache.hadoop.fs.FileStatus
9: 676141 203208736 [B
10: 3845164 184567872 java.util.Hashtable$Entry
11: 3174033 152353584 java.util.concurrent.ConcurrentHashMap$Node
```
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Added UT
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] [spark] mridulm commented on pull request #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
Posted by "mridulm (via GitHub)" <gi...@apache.org>.
mridulm commented on PR #41676:
URL: https://github.com/apache/spark/pull/41676#issuecomment-1614239102
> Hi, @mridulm Thanks for your reminder. Can you describe the non-shuffle scenario more specifically?
It is not very common for shuffle stages to have preferred locality (unless push based - where node is always 1 or nontrivial fraction of reducer input generated in an executor).
It is much more common for non-shuffle stages to surface locality.
Depending on the underlying impl, this can or need not have duplication: hdfs for example does not have duplication, while replicated RDD could, and so on.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] [spark] dongjoon-hyun commented on a diff in pull request #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
Posted by "dongjoon-hyun (via GitHub)" <gi...@apache.org>.
dongjoon-hyun commented on code in PR #41676:
URL: https://github.com/apache/spark/pull/41676#discussion_r1235622164
##########
core/src/main/scala/org/apache/spark/MapOutputTracker.scala:
##########
@@ -1122,7 +1122,7 @@ private[spark] class MapOutputTrackerMaster(
if (startMapIndex < endMapIndex &&
(startMapIndex >= 0 && endMapIndex <= statuses.length)) {
val statusesPicked = statuses.slice(startMapIndex, endMapIndex).filter(_ != null)
- statusesPicked.map(_.location.host).toSeq
+ statusesPicked.map(_.location.host).toSeq.distinct
Review Comment:
+1, for @LuciferYang 's comment.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] [spark] dongjoon-hyun commented on pull request #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
Posted by "dongjoon-hyun (via GitHub)" <gi...@apache.org>.
dongjoon-hyun commented on PR #41676:
URL: https://github.com/apache/spark/pull/41676#issuecomment-1600951488
Merged to master. Thank you, @wankunde , @LuciferYang and @wangyum .
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] [spark] wankunde commented on pull request #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
Posted by "wankunde (via GitHub)" <gi...@apache.org>.
wankunde commented on PR #41676:
URL: https://github.com/apache/spark/pull/41676#issuecomment-1600083560
Retest this please
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] [spark] dongjoon-hyun commented on pull request #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
Posted by "dongjoon-hyun (via GitHub)" <gi...@apache.org>.
dongjoon-hyun commented on PR #41676:
URL: https://github.com/apache/spark/pull/41676#issuecomment-1599983267
Hi, @wankunde . Could you take a look at your GitHub Action pipeline? One job seems to be stuck.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] [spark] LuciferYang commented on a diff in pull request #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
Posted by "LuciferYang (via GitHub)" <gi...@apache.org>.
LuciferYang commented on code in PR #41676:
URL: https://github.com/apache/spark/pull/41676#discussion_r1235612021
##########
core/src/main/scala/org/apache/spark/MapOutputTracker.scala:
##########
@@ -1122,7 +1122,7 @@ private[spark] class MapOutputTrackerMaster(
if (startMapIndex < endMapIndex &&
(startMapIndex >= 0 && endMapIndex <= statuses.length)) {
val statusesPicked = statuses.slice(startMapIndex, endMapIndex).filter(_ != null)
- statusesPicked.map(_.location.host).toSeq
+ statusesPicked.map(_.location.host).toSeq.distinct
Review Comment:
hmm... Is `.distinct.toSeq` more friendly when using Scala 2.13?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] [spark] wankunde commented on a diff in pull request #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
Posted by "wankunde (via GitHub)" <gi...@apache.org>.
wankunde commented on code in PR #41676:
URL: https://github.com/apache/spark/pull/41676#discussion_r1235800177
##########
core/src/main/scala/org/apache/spark/MapOutputTracker.scala:
##########
@@ -1122,7 +1122,7 @@ private[spark] class MapOutputTrackerMaster(
if (startMapIndex < endMapIndex &&
(startMapIndex >= 0 && endMapIndex <= statuses.length)) {
val statusesPicked = statuses.slice(startMapIndex, endMapIndex).filter(_ != null)
- statusesPicked.map(_.location.host).toSeq
+ statusesPicked.map(_.location.host).toSeq.distinct
Review Comment:
Thanks.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] [spark] wankunde commented on pull request #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
Posted by "wankunde (via GitHub)" <gi...@apache.org>.
wankunde commented on PR #41676:
URL: https://github.com/apache/spark/pull/41676#issuecomment-1612559700
> Not handling this for non-shuffle ? Shuffle actually has limited oppurtunities for pref locality compared to the rest.
Hi, @mridulm Thanks for your reminder. Can you describe the non-shuffle scenario more specifically?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] [spark] dongjoon-hyun closed pull request #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
Posted by "dongjoon-hyun (via GitHub)" <gi...@apache.org>.
dongjoon-hyun closed pull request #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
URL: https://github.com/apache/spark/pull/41676
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
[GitHub] [spark] mridulm commented on pull request #41676: [SPARK-44109][CORE] Remove duplicate preferred locations of each RDD partition
Posted by "mridulm (via GitHub)" <gi...@apache.org>.
mridulm commented on PR #41676:
URL: https://github.com/apache/spark/pull/41676#issuecomment-1605579525
Not handling this for shuffle ?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org