You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@spark.apache.org by "sebastianhillig-db (via GitHub)" <gi...@apache.org> on 2024/03/21 09:16:09 UTC
[PR] [WIP] First stab at dealing with worker crashes [spark]
sebastianhillig-db opened a new pull request, #45635:
URL: https://github.com/apache/spark/pull/45635
<!--
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
'common/utils/src/main/resources/error/README.md'.
-->
### What changes were proposed in this pull request?
<!--
Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster reviews in your PR. See the examples below.
1. If you refactor some codes with changing classes, showing the class hierarchy will help reviewers.
2. If you fix some SQL features, you can provide some references of other DBMSes.
3. If there is design documentation, please add the link.
4. If there is a discussion in the mailing list, please add the link.
-->
### Why are the changes needed?
<!--
Please clarify why the changes are needed. For instance,
1. If you propose a new API, clarify the use case for a new API.
2. If you fix a bug, you can clarify why it is a bug.
-->
### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such as the documentation fix.
If yes, please clarify the previous behavior and the change this PR proposes - provide the console output, description and/or an example to show the behavior difference if possible.
If possible, please also clarify if this is a user-facing change compared to the released Spark versions or within the unreleased branches such as master.
If no, write 'No'.
-->
### How was this patch tested?
<!--
If tests were added, say they were added here. Please make sure to add some test cases that check the changes thoroughly including negative and positive cases if possible.
If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future.
If tests were not added, please describe why they were not added and/or why it was difficult to add.
If benchmark tests were added, please run the benchmarks in GitHub Actions for the consistent environment, and the instructions could accord to: https://spark.apache.org/developer-tools.html#github-workflow-benchmarks.
-->
### Was this patch authored or co-authored using generative AI tooling?
<!--
If generative AI tooling has been used in the process of authoring this patch, please include the
phrase: 'Generated-by: ' followed by the name of the tool and its version.
If no, write 'No'.
Please refer to the [ASF Generative Tooling Guidance](https://www.apache.org/legal/generative-tooling.html) for details.
-->
--
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
Re: [PR] [WIP] First stab at dealing with worker crashes while idling [spark]
Posted by "utkarsh39 (via GitHub)" <gi...@apache.org>.
utkarsh39 commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1535780432
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,19 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ while (idleWorkers.nonEmpty) {
val worker = idleWorkers.dequeue()
- worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
- return (worker, daemonWorkers.get(worker))
+ val workerHandle = daemonWorkers(worker)
+ try {
+ worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
+ } catch {
+ case c: CancelledKeyException => /* pass */
+ }
+ if (workerHandle.isAlive() && worker.selectionKey.isValid()) {
Review Comment:
If the `interestOps` call succeed, will both of these checks be automatically true?
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,19 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ while (idleWorkers.nonEmpty) {
val worker = idleWorkers.dequeue()
- worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
- return (worker, daemonWorkers.get(worker))
+ val workerHandle = daemonWorkers(worker)
+ try {
+ worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
+ } catch {
+ case c: CancelledKeyException => /* pass */
+ }
+ if (workerHandle.isAlive() && worker.selectionKey.isValid()) {
+ return (worker, Some(workerHandle.pid()))
Review Comment:
Should we move this to the try block?
##########
python/pyspark/tests/test_worker.py:
##########
@@ -256,6 +257,20 @@ def conf(cls):
return _conf
+class WorkerPoolCrashTest(PySparkTestCase):
+ def test_worker_crash(self):
Review Comment:
Does this test fail today?
--
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
Re: [PR] [SPARK-47565][PYTHON] PySpark worker pool crash resilience [spark]
Posted by "sebastianhillig-db (via GitHub)" <gi...@apache.org>.
sebastianhillig-db commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1542041741
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,20 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ // Pull from idle workers until we one that is alive, otherwise create a new one.
+ while (idleWorkers.nonEmpty) {
Review Comment:
Ugh, sorry - the force push broke that link. I'm referring to "releaseWorker" using the same synchronization, so we should not be adding new workers while this code runs.
--
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
Re: [PR] [SPARK-47565][PYTHON] PySpark worker pool crash resilience [spark]
Posted by "sebastianhillig-db (via GitHub)" <gi...@apache.org>.
sebastianhillig-db commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1542043094
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,20 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ // Pull from idle workers until we one that is alive, otherwise create a new one.
+ while (idleWorkers.nonEmpty) {
val worker = idleWorkers.dequeue()
- worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
- return (worker, daemonWorkers.get(worker))
+ val workerHandle = daemonWorkers(worker)
Review Comment:
Since the idleWorkers queue keeps the reference alive, there should never be a case where the daemonWorker disappears until it is retrieved.
--
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
Re: [PR] [WIP] First stab at dealing with worker crashes while idling [spark]
Posted by "sebastianhillig-db (via GitHub)" <gi...@apache.org>.
sebastianhillig-db commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1535787390
##########
python/pyspark/tests/test_worker.py:
##########
@@ -256,6 +257,20 @@ def conf(cls):
return _conf
+class WorkerPoolCrashTest(PySparkTestCase):
+ def test_worker_crash(self):
Review Comment:
Yes!
--
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
Re: [PR] [WIP] First stab at dealing with worker crashes while idling [spark]
Posted by "utkarsh39 (via GitHub)" <gi...@apache.org>.
utkarsh39 commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1535799832
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,19 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
Review Comment:
Do you know what happens in the case where the worker dies right after the `create()` call returns successfully? I think we are rightfully solving one of the cases here where an idle worker sitting in the `idleWorkers` has died. The other case can still happen
--
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
Re: [PR] PySpark worker pool crash resilience [spark]
Posted by "HyukjinKwon (via GitHub)" <gi...@apache.org>.
HyukjinKwon commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1538459016
##########
python/pyspark/tests/test_worker.py:
##########
@@ -256,6 +257,20 @@ def conf(cls):
return _conf
+class WorkerPoolCrashTest(PySparkTestCase):
+ def test_worker_crash(self):
Review Comment:
Should add a comment like:
```
def test_case(self):
# SPARK-12345: a short description of the test
...
```
--
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
Re: [PR] [SPARK-47565][PYTHON] PySpark worker pool crash resilience [spark]
Posted by "dongjoon-hyun (via GitHub)" <gi...@apache.org>.
dongjoon-hyun commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1541298473
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,20 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ // Pull from idle workers until we one that is alive, otherwise create a new one.
+ while (idleWorkers.nonEmpty) {
Review Comment:
> On each iteration, a worker is pulled from `idleWorkers`, this will end up "emptying" the pool. The synchronization around this will ensure that no other workers are added while this happens. (see https://github.com/apache/spark/pull/45635/files/ba3c6f6ee19762278004594735f25ab4f6fafb3e#diff-1bd846874b06327e6abd0803aa74eed890352dfa974d5c1da1a12dc7477e20d0L411-L413)
The link seems to be broken.
--
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
Re: [PR] [SPARK-47565][PYTHON] PySpark worker pool crash resilience [spark]
Posted by "ueshin (via GitHub)" <gi...@apache.org>.
ueshin commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1541986448
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,20 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ // Pull from idle workers until we one that is alive, otherwise create a new one.
+ while (idleWorkers.nonEmpty) {
val worker = idleWorkers.dequeue()
- worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
- return (worker, daemonWorkers.get(worker))
+ val workerHandle = daemonWorkers(worker)
Review Comment:
Do we have a chance that this throws an exception?
If so, should we use `daemonWorkers.get(worker)` same as before, just in case?
--
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
Re: [PR] [SPARK-47565] PySpark worker pool crash resilience [spark]
Posted by "dongjoon-hyun (via GitHub)" <gi...@apache.org>.
dongjoon-hyun commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1539738583
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,20 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ // Pull from idle workers until we one that is alive, otherwise create a new one.
+ while (idleWorkers.nonEmpty) {
Review Comment:
It seems that there is a chance to introduce an infinite loop to Apache Spark. Maybe, limit the number of retry? WDYT, @sebastianhillig-db ?
--
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
Re: [PR] [WIP] First stab at dealing with worker crashes while idling [spark]
Posted by "sebastianhillig-db (via GitHub)" <gi...@apache.org>.
sebastianhillig-db commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1535790816
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,19 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ while (idleWorkers.nonEmpty) {
val worker = idleWorkers.dequeue()
- worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
- return (worker, daemonWorkers.get(worker))
+ val workerHandle = daemonWorkers(worker)
+ try {
+ worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
+ } catch {
+ case c: CancelledKeyException => /* pass */
+ }
+ if (workerHandle.isAlive() && worker.selectionKey.isValid()) {
+ return (worker, Some(workerHandle.pid()))
Review Comment:
We could reorder it like this?
```
val workerHandle = daemonWorkers(worker)
if (workerHandle.isAlive()
try {
worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
return (worker, Some(workerHandle.pid()))
} catch {
case c: CancelledKeyException => /* pass */
}
}
logWarning(s"Worker ${worker} process from idle queue is dead, discarding.")
```
--
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
Re: [PR] [WIP] First stab at dealing with worker crashes while idling [spark]
Posted by "utkarsh39 (via GitHub)" <gi...@apache.org>.
utkarsh39 commented on PR #45635:
URL: https://github.com/apache/spark/pull/45635#issuecomment-2015554167
LGTM. Let's get a review from others?
--
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
Re: [PR] [WIP] First stab at dealing with worker crashes while idling [spark]
Posted by "sebastianhillig-db (via GitHub)" <gi...@apache.org>.
sebastianhillig-db commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1535805036
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,19 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
Review Comment:
There's a bit of a timing question there, because the factory does not wait for a first bits to be sent to the worker. So this would result in an IO error still. We could add a first bit to be sent by the factory to the forked of process to signal aliveness though to capture this?
--
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
Re: [PR] [WIP] First stab at dealing with worker crashes while idling [spark]
Posted by "sebastianhillig-db (via GitHub)" <gi...@apache.org>.
sebastianhillig-db commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1535790816
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,19 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ while (idleWorkers.nonEmpty) {
val worker = idleWorkers.dequeue()
- worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
- return (worker, daemonWorkers.get(worker))
+ val workerHandle = daemonWorkers(worker)
+ try {
+ worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
+ } catch {
+ case c: CancelledKeyException => /* pass */
+ }
+ if (workerHandle.isAlive() && worker.selectionKey.isValid()) {
+ return (worker, Some(workerHandle.pid()))
Review Comment:
We could reorder it like this?
```scala
val workerHandle = daemonWorkers(worker)
if (workerHandle.isAlive()) {
try {
worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
return (worker, Some(workerHandle.pid()))
} catch {
case c: CancelledKeyException => /* pass */
}
}
logWarning(s"Worker ${worker} process from idle queue is dead, discarding.")
```
--
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
Re: [PR] [SPARK-47565][PYTHON] PySpark worker pool crash resilience [spark]
Posted by "HyukjinKwon (via GitHub)" <gi...@apache.org>.
HyukjinKwon closed pull request #45635: [SPARK-47565][PYTHON] PySpark worker pool crash resilience
URL: https://github.com/apache/spark/pull/45635
--
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
Re: [PR] [SPARK-47565][PYTHON] PySpark worker pool crash resilience [spark]
Posted by "HyukjinKwon (via GitHub)" <gi...@apache.org>.
HyukjinKwon commented on PR #45635:
URL: https://github.com/apache/spark/pull/45635#issuecomment-2036882209
Merged to master.
--
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
Re: [PR] PySpark worker pool crash resilience [spark]
Posted by "HyukjinKwon (via GitHub)" <gi...@apache.org>.
HyukjinKwon commented on PR #45635:
URL: https://github.com/apache/spark/pull/45635#issuecomment-2019204500
Apache Spark uses the GitHub Actions in your forked repository so the builds have to be found in https://github.com/sebastianhillig-db/spark/actions . The GitHub Actions would have to be enabled at https://github.com/sebastianhillig-db/spark/settings/actions , and rebase this PR
--
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
Re: [PR] [SPARK-47565] PySpark worker pool crash resilience [spark]
Posted by "sebastianhillig-db (via GitHub)" <gi...@apache.org>.
sebastianhillig-db commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1539138374
##########
python/pyspark/tests/test_worker.py:
##########
@@ -256,6 +257,20 @@ def conf(cls):
return _conf
+class WorkerPoolCrashTest(PySparkTestCase):
+ def test_worker_crash(self):
Review Comment:
Done!
--
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
Re: [PR] [SPARK-47565][PYTHON] PySpark worker pool crash resilience [spark]
Posted by "sebastianhillig-db (via GitHub)" <gi...@apache.org>.
sebastianhillig-db commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1540623063
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,20 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ // Pull from idle workers until we one that is alive, otherwise create a new one.
+ while (idleWorkers.nonEmpty) {
Review Comment:
On each iteration, a worker is pulled from `idleWorkers`, this will end up "emptying" the pool. The synchronization around this will ensure that no other workers are added while this happens. (see https://github.com/apache/spark/pull/45635/files/ba3c6f6ee19762278004594735f25ab4f6fafb3e#diff-1bd846874b06327e6abd0803aa74eed890352dfa974d5c1da1a12dc7477e20d0L411-L413)
--
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
Re: [PR] [WIP] First stab at dealing with worker crashes while idling [spark]
Posted by "utkarsh39 (via GitHub)" <gi...@apache.org>.
utkarsh39 commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1535852476
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,19 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
Review Comment:
I see. I think that might be an overkill. This is good enough IMO
--
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
Re: [PR] [WIP] First stab at dealing with worker crashes while idling [spark]
Posted by "sebastianhillig-db (via GitHub)" <gi...@apache.org>.
sebastianhillig-db commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1535787105
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,19 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ while (idleWorkers.nonEmpty) {
val worker = idleWorkers.dequeue()
- worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
- return (worker, daemonWorkers.get(worker))
+ val workerHandle = daemonWorkers(worker)
+ try {
+ worker.selectionKey.interestOps(SelectionKey.OP_READ | SelectionKey.OP_WRITE)
+ } catch {
+ case c: CancelledKeyException => /* pass */
+ }
+ if (workerHandle.isAlive() && worker.selectionKey.isValid()) {
Review Comment:
It seems that this isn't always the case, i.e. the workerHandle may already see the process being dead and selectionKey update will happily pass. I also check `isValid` for the off-chance, that we got cancelled after interestOps was set.
--
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
Re: [PR] PySpark worker pool crash resilience [spark]
Posted by "sebastianhillig-db (via GitHub)" <gi...@apache.org>.
sebastianhillig-db commented on PR #45635:
URL: https://github.com/apache/spark/pull/45635#issuecomment-2018302912
@ueshin @HyukjinKwon can you take a look here?
--
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
Re: [PR] PySpark worker pool crash resilience [spark]
Posted by "HyukjinKwon (via GitHub)" <gi...@apache.org>.
HyukjinKwon commented on PR #45635:
URL: https://github.com/apache/spark/pull/45635#issuecomment-2019202983
Let's file a JIRA, see https://spark.apache.org/contributing.html
--
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
Re: [PR] [SPARK-47565][PYTHON] PySpark worker pool crash resilience [spark]
Posted by "sebastianhillig-db (via GitHub)" <gi...@apache.org>.
sebastianhillig-db commented on code in PR #45635:
URL: https://github.com/apache/spark/pull/45635#discussion_r1542043889
##########
core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala:
##########
@@ -95,10 +95,20 @@ private[spark] class PythonWorkerFactory(
def create(): (PythonWorker, Option[Long]) = {
if (useDaemon) {
self.synchronized {
- if (idleWorkers.nonEmpty) {
+ // Pull from idle workers until we one that is alive, otherwise create a new one.
+ while (idleWorkers.nonEmpty) {
Review Comment:
Promise not to force push again: https://github.com/apache/spark/pull/45635/files#diff-1bd846874b06327e6abd0803aa74eed890352dfa974d5c1da1a12dc7477e20d0L411
--
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