You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by ChengzhiZhao <gi...@git.apache.org> on 2018/02/18 21:10:05 UTC
[GitHub] flink pull request #5521: [FLINK-8599] Improve the failure behavior of the F...
GitHub user ChengzhiZhao opened a pull request:
https://github.com/apache/flink/pull/5521
[FLINK-8599] Improve the failure behavior of the FileInputFormat for …
## What is the purpose of the change
This pull request is intent to improve the failure behavior of the ContinuousFileReader, currently if a bad file (for example, a different schema been dropped in this folder) came to the path and flink will do several retries.
However, since the file path persist in the checkpoint, when people tried to resume from external checkpoint, it threw the following error on no file been found and the process cannot move forward.
`java.io.IOException: Error opening the Input Split s3a://myfile [0,904]: No such file or directory: s3a://myfile`
The change is to check if the path exist before open the file, if error occurs and bad file removed, flink should resume the process and continue.
## Brief change log
- *Add a file exist check before open the file *
## Verifying this change
- *Manually verified the change by introduce a bad file while continuously monitoring the folder, after remove the bad file, the process continued.*
## Does this pull request potentially affect one of the following parts:
- Dependencies (does it add or upgrade a dependency): (no)
- The public API, i.e., is any changed class annotated with `@Public(Evolving)`: (no)
- The serializers: (no)
- The runtime per-record code paths (performance sensitive): (no)
- Anything that affects deployment or recovery: JobManager (and its components), Checkpointing, Yarn/Mesos, ZooKeeper: (no)
- The S3 file system connector: (no)
## Documentation
- Does this pull request introduce a new feature? (no)
- If yes, how is the feature documented? (not applicable)
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/ChengzhiZhao/flink Improve_failure_behavior_FileInputFormat
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/flink/pull/5521.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #5521
----
commit 6fa8ef212c536acee56b2e9831ec92d1059449ff
Author: Chengzhi Zhao <ch...@...>
Date: 2018-02-18T18:23:32Z
[FLINK-8599] Improve the failure behavior of the FileInputFormat for bad files
----
---
[GitHub] flink pull request #5521: [FLINK-8599] Improve the failure behavior of the F...
Posted by steveloughran <gi...@git.apache.org>.
Github user steveloughran commented on a diff in the pull request:
https://github.com/apache/flink/pull/5521#discussion_r169275364
--- Diff: flink-core/src/main/java/org/apache/flink/api/common/io/FileInputFormat.java ---
@@ -706,6 +700,9 @@ public void open(FileInputSplit fileSplit) throws IOException {
this.stream = isot.waitForCompletion();
this.stream = decorateInputStream(this.stream, fileSplit);
}
+ catch (FileNotFoundException e) {
+ throw new FileNotFoundException("Input split " + fileSplit.getPath() + " doesn't exist, skip and continue");
+ }
--- End diff --
I would recommend including the text and stack of the caught ex, for the better stack trace. FNFEs can get raised in odd circumstances in S3; seeing the full stack is what you need when fielding support calls. eg.
```java
throw new FileNotFoundException("Input split " + fileSplit.getPath() + " doesn't exist, skip and continue: " + e, e);
```
---
[GitHub] flink issue #5521: [FLINK-8599] Improve the failure behavior of the FileInpu...
Posted by StephanEwen <gi...@git.apache.org>.
Github user StephanEwen commented on the issue:
https://github.com/apache/flink/pull/5521
The approach here does not work (there is also no test, otherwise that would have been caught).
Flink does not magically load configs like Hadoop, Flink passes configs explicitly.
Aside from that, we first should actually decide whether this should be
---
[GitHub] flink pull request #5521: [FLINK-8599] Improve the failure behavior of the F...
Posted by steveloughran <gi...@git.apache.org>.
Github user steveloughran commented on a diff in the pull request:
https://github.com/apache/flink/pull/5521#discussion_r169604725
--- Diff: flink-core/src/main/java/org/apache/flink/api/common/io/FileInputFormat.java ---
@@ -819,6 +819,10 @@ public void open(FileInputSplit fileSplit) throws IOException {
this.stream = isot.waitForCompletion();
this.stream = decorateInputStream(this.stream, fileSplit);
}
+ catch (FileNotFoundException e) {
+ throw new FileNotFoundException("Input split " + fileSplit.getPath() +
+ " doesn't exist, skip and continue: " + e.getMessage());
+ }
--- End diff --
As this exception doesn't have a constructor which takes a nested exception, can you use `initCause()` to patch it.
```java
throw (FileNotFoundException)(new FileNotFoundException(...)).initCause(e)
```
---
[GitHub] flink pull request #5521: [FLINK-8599] Improve the failure behavior of the F...
Posted by steveloughran <gi...@git.apache.org>.
Github user steveloughran commented on a diff in the pull request:
https://github.com/apache/flink/pull/5521#discussion_r169132404
--- Diff: flink-core/src/main/java/org/apache/flink/api/common/io/FileInputFormat.java ---
@@ -691,6 +691,12 @@ public void open(FileInputSplit fileSplit) throws IOException {
LOG.debug("Opening input split " + fileSplit.getPath() + " [" + this.splitStart + "," + this.splitLength + "]");
}
+ if (!exists(fileSplit.getPath())) {
--- End diff --
you are doubling the number of checks for file existence here, which, when working with S3 implies three more HTTP requests which takes time and cost money. Better to do the open() call and catch FileNotFoundException, which all filesystems are required to throw if they are given a path which doesn't resolve to a file.
---
[GitHub] flink issue #5521: [FLINK-8599] Improve the failure behavior of the FileInpu...
Posted by ChengzhiZhao <gi...@git.apache.org>.
Github user ChengzhiZhao commented on the issue:
https://github.com/apache/flink/pull/5521
Thanks @steveloughran for your feedbacks, I updated based on your suggestions.
---
[GitHub] flink issue #5521: [FLINK-8599] Improve the failure behavior of the FileInpu...
Posted by ChengzhiZhao <gi...@git.apache.org>.
Github user ChengzhiZhao commented on the issue:
https://github.com/apache/flink/pull/5521
@StephanEwen, thanks for your feedbacks. for our use case we'd like to take the bad file out to continue process instead of fixing the bad file to be able to continue. I think it can be more as job/application specific so we won't have it globally, it will give people flexibility. If that is the decision here, we can do as a connector parameter not via config which affect all the jobs. Please suggest.
---
[GitHub] flink issue #5521: [FLINK-8599] Improve the failure behavior of the FileInpu...
Posted by kl0u <gi...@git.apache.org>.
Github user kl0u commented on the issue:
https://github.com/apache/flink/pull/5521
@ChengzhiZhao I think that we should handle the "bad" splits better.
@StephanEwen You are right that we should give the user the option to fail the job or ignore them.
---
[GitHub] flink issue #5521: [FLINK-8599] Improve the failure behavior of the FileInpu...
Posted by ChengzhiZhao <gi...@git.apache.org>.
Github user ChengzhiZhao commented on the issue:
https://github.com/apache/flink/pull/5521
@steveloughran Updated.
---
[GitHub] flink issue #5521: [FLINK-8599] Improve the failure behavior of the FileInpu...
Posted by ChengzhiZhao <gi...@git.apache.org>.
Github user ChengzhiZhao commented on the issue:
https://github.com/apache/flink/pull/5521
@StephanEwen @kl0u Thanks for you feedback, I will put an option for user to choose
---
[GitHub] flink pull request #5521: [FLINK-8599] Improve the failure behavior of the F...
Posted by StephanEwen <gi...@git.apache.org>.
Github user StephanEwen commented on a diff in the pull request:
https://github.com/apache/flink/pull/5521#discussion_r173454031
--- Diff: flink-core/src/main/java/org/apache/flink/api/common/io/FileInputFormat.java ---
@@ -819,6 +819,10 @@ public void open(FileInputSplit fileSplit) throws IOException {
this.stream = isot.waitForCompletion();
this.stream = decorateInputStream(this.stream, fileSplit);
}
+ catch (FileNotFoundException e) {
+ throw (FileNotFoundException)(new FileNotFoundException("Input split " + fileSplit.getPath() +
--- End diff --
I don't understand why "skip and continue" is in this message.
Not all users of the `FileIputFormat` skip and continue. The interpretation of the exception should not be assumed when creating the exception.
---
[GitHub] flink pull request #5521: [FLINK-8599] Improve the failure behavior of the F...
Posted by StephanEwen <gi...@git.apache.org>.
Github user StephanEwen commented on a diff in the pull request:
https://github.com/apache/flink/pull/5521#discussion_r173453379
--- Diff: flink-streaming-java/src/main/java/org/apache/flink/streaming/api/functions/source/ContinuousFileReaderOperator.java ---
@@ -340,6 +341,10 @@ public void run() {
}
}
+ } catch (FileNotFoundException e) {
+ if (LOG.isDebugEnabled()) {
--- End diff --
This needs to be more prominently logged than `debug`. Should be at least `info` or `warn`.
Please also use the placeholder syntax from log4j, to make the code simpler.
---
[GitHub] flink issue #5521: [FLINK-8599] Improve the failure behavior of the FileInpu...
Posted by steveloughran <gi...@git.apache.org>.
Github user steveloughran commented on the issue:
https://github.com/apache/flink/pull/5521
This is very inefficient against an object store, potentially adding a few hundred millis and $0.01 per file. I would simply catch FileNotFoundExceptions raised in the open() call and treat them specially
See [How long does FileSystem.exists() take against S3?](http://steveloughran.blogspot.co.uk/2016/12/how-long-does-filesystemexists-take.html)
---