[
https://issues.apache.org/jira/browse/FLINK-25478?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yun Tang updated FLINK-25478:
-----------------------------
Description:
Currently, changelog materialization would call RocksDB state backend's
snapshot method to generate {{IncrementalRemoteKeyedStateHandle}} as
ChangelogStateBackendHandleImpl's materialized artifacts. And before next
materialization, it will always report the same
{{IncrementalRemoteKeyedStateHandle}} as before.
It's fine to register this for the 1st time. However, for the 2nd time to
register {{IncrementalRemoteKeyedStateHandle}} (via
{{{}ChangelogStateBackendHandleImpl#registerSharedStates{}}}), it will discard
the private state artifacts without check the register reference:
IncrementalRemoteKeyedStateHandle:
{code:java}
public void discardState() throws Exception {
try {
StateUtil.bestEffortDiscardAllStateObjects(privateState.values());
} catch (Exception e) {
LOG.warn("Could not properly discard misc file states.", e);
}
}
{code}
Thus, this would delete the private state (such as RocksDB's MAINFEST), and
once restore, job would not report FileNotFoundException.
Caused by: org.apache.flink.runtime.state.BackendBuildingException: Caught
unexpected exception.
at
org.apache.flink.contrib.streaming.state.RocksDBKeyedStateBackendBuilder.build(RocksDBKeyedStateBackendBuilder.java:403)
at
org.apache.flink.contrib.streaming.state.EmbeddedRocksDBStateBackend.createKeyedStateBackend(EmbeddedRocksDBStateBackend.java:477)
at
org.apache.flink.contrib.streaming.state.EmbeddedRocksDBStateBackend.createKeyedStateBackend(EmbeddedRocksDBStateBackend.java:90)
at
org.apache.flink.state.changelog.ChangelogStateBackend.lambda$createKeyedStateBackend$1(ChangelogStateBackend.java:153)
at
org.apache.flink.state.changelog.restore.ChangelogBackendRestoreOperation.restore(ChangelogBackendRestoreOperation.java:68)
at
org.apache.flink.state.changelog.ChangelogStateBackend.restore(ChangelogStateBackend.java:221)
at
org.apache.flink.state.changelog.ChangelogStateBackend.createKeyedStateBackend(ChangelogStateBackend.java:145)
at
org.apache.flink.streaming.api.operators.StreamTaskStateInitializerImpl.lambda$keyedStatedBackend$1(StreamTaskStateInitializerImpl.java:328)
at
org.apache.flink.streaming.api.operators.BackendRestorerProcedure.attemptCreateAndRestore(BackendRestorerProcedure.java:168)
at
org.apache.flink.streaming.api.operators.BackendRestorerProcedure.createAndRestore(BackendRestorerProcedure.java:135)
at
org.apache.flink.streaming.api.operators.StreamTaskStateInitializerImpl.keyedStatedBackend(StreamTaskStateInitializerImpl.java:345)
at
org.apache.flink.streaming.api.operators.StreamTaskStateInitializerImpl.streamOperatorStateContext(StreamTaskStateInitializerImpl.java:163)
... 10 more
Caused by: java.io.FileNotFoundException: xxxxx
at
org.apache.flink.core.fs.PluginFileSystemFactory$ClassLoaderFixingFileSystem.open(PluginFileSystemFactory.java:127)
at
org.apache.flink.runtime.state.filesystem.FileStateHandle.openInputStream(FileStateHandle.java:68)
at
org.apache.flink.contrib.streaming.state.RocksDBStateDownloader.downloadDataForStateHandle(RocksDBStateDownloader.java:127)
at
org.apache.flink.contrib.streaming.state.RocksDBStateDownloader.lambda$createDownloadRunnables$0(RocksDBStateDownloader.java:110)
at
org.apache.flink.util.function.ThrowingRunnable.lambda$unchecked$0(ThrowingRunnable.java:49)
at
java.util.concurrent.CompletableFuture$AsyncRun.run(CompletableFuture.java:1626)
~[?:1.8.0_102]
at
java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1147)
~[?:1.8.0_102]
at
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:622)
~[?:1.8.0_102]
... 1 more
was:
Currently, changelog materialization would call RocksDB state backend's
snapshot method to generate {{IncrementalRemoteKeyedStateHandle}} as
ChangelogStateBackendHandleImpl's materialized artifacts. And before next
materialization, it will always report the same
{{IncrementalRemoteKeyedStateHandle}} as before.
It's fine to register this for the 1st time. However, for the 2nd time to
register {{IncrementalRemoteKeyedStateHandle}} (via
{{ChangelogStateBackendHandleImpl#registerSharedStates}}), it will discard the
private state artifacts without check the register reference:
IncrementalRemoteKeyedStateHandle:
{code:java}
public void discardState() throws Exception {
try {
StateUtil.bestEffortDiscardAllStateObjects(privateState.values());
} catch (Exception e) {
LOG.warn("Could not properly discard misc file states.", e);
}
}
{code}
Thus, this would delete the private state (such as RocksDB's MAINFEST), and
once restore, job would not report FileNotFoundException.
> Changelog materialization with incremental checkpoint could cause
> checkpointed data lost
> ----------------------------------------------------------------------------------------
>
> Key: FLINK-25478
> URL: https://issues.apache.org/jira/browse/FLINK-25478
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Checkpointing, Runtime / State Backends
> Reporter: Yun Tang
> Priority: Critical
> Fix For: 1.15.0
>
>
> Currently, changelog materialization would call RocksDB state backend's
> snapshot method to generate {{IncrementalRemoteKeyedStateHandle}} as
> ChangelogStateBackendHandleImpl's materialized artifacts. And before next
> materialization, it will always report the same
> {{IncrementalRemoteKeyedStateHandle}} as before.
> It's fine to register this for the 1st time. However, for the 2nd time to
> register {{IncrementalRemoteKeyedStateHandle}} (via
> {{{}ChangelogStateBackendHandleImpl#registerSharedStates{}}}), it will
> discard the private state artifacts without check the register reference:
> IncrementalRemoteKeyedStateHandle:
> {code:java}
> public void discardState() throws Exception {
> try {
> StateUtil.bestEffortDiscardAllStateObjects(privateState.values());
> } catch (Exception e) {
> LOG.warn("Could not properly discard misc file states.", e);
> }
> }
> {code}
> Thus, this would delete the private state (such as RocksDB's MAINFEST), and
> once restore, job would not report FileNotFoundException.
>
> Caused by: org.apache.flink.runtime.state.BackendBuildingException: Caught
> unexpected exception.
> at
> org.apache.flink.contrib.streaming.state.RocksDBKeyedStateBackendBuilder.build(RocksDBKeyedStateBackendBuilder.java:403)
> at
> org.apache.flink.contrib.streaming.state.EmbeddedRocksDBStateBackend.createKeyedStateBackend(EmbeddedRocksDBStateBackend.java:477)
>
> at
> org.apache.flink.contrib.streaming.state.EmbeddedRocksDBStateBackend.createKeyedStateBackend(EmbeddedRocksDBStateBackend.java:90)
>
> at
> org.apache.flink.state.changelog.ChangelogStateBackend.lambda$createKeyedStateBackend$1(ChangelogStateBackend.java:153)
> at
> org.apache.flink.state.changelog.restore.ChangelogBackendRestoreOperation.restore(ChangelogBackendRestoreOperation.java:68)
> at
> org.apache.flink.state.changelog.ChangelogStateBackend.restore(ChangelogStateBackend.java:221)
>
> at
> org.apache.flink.state.changelog.ChangelogStateBackend.createKeyedStateBackend(ChangelogStateBackend.java:145)
>
> at
> org.apache.flink.streaming.api.operators.StreamTaskStateInitializerImpl.lambda$keyedStatedBackend$1(StreamTaskStateInitializerImpl.java:328)
>
> at
> org.apache.flink.streaming.api.operators.BackendRestorerProcedure.attemptCreateAndRestore(BackendRestorerProcedure.java:168)
>
> at
> org.apache.flink.streaming.api.operators.BackendRestorerProcedure.createAndRestore(BackendRestorerProcedure.java:135)
>
> at
> org.apache.flink.streaming.api.operators.StreamTaskStateInitializerImpl.keyedStatedBackend(StreamTaskStateInitializerImpl.java:345)
>
> at
> org.apache.flink.streaming.api.operators.StreamTaskStateInitializerImpl.streamOperatorStateContext(StreamTaskStateInitializerImpl.java:163)
>
> ... 10 more
> Caused by: java.io.FileNotFoundException: xxxxx
> at
> org.apache.flink.core.fs.PluginFileSystemFactory$ClassLoaderFixingFileSystem.open(PluginFileSystemFactory.java:127)
>
> at
> org.apache.flink.runtime.state.filesystem.FileStateHandle.openInputStream(FileStateHandle.java:68)
>
> at
> org.apache.flink.contrib.streaming.state.RocksDBStateDownloader.downloadDataForStateHandle(RocksDBStateDownloader.java:127)
>
> at
> org.apache.flink.contrib.streaming.state.RocksDBStateDownloader.lambda$createDownloadRunnables$0(RocksDBStateDownloader.java:110)
> at
> org.apache.flink.util.function.ThrowingRunnable.lambda$unchecked$0(ThrowingRunnable.java:49)
>
> at
> java.util.concurrent.CompletableFuture$AsyncRun.run(CompletableFuture.java:1626)
> ~[?:1.8.0_102]
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1147)
> ~[?:1.8.0_102]
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:622)
> ~[?:1.8.0_102]
> ... 1 more
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
