Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[JENKINS-55287] Improve error messages for nonresumable Pipelines #363

Closed
wants to merge 10 commits into from
Closed

[JENKINS-55287] Improve error messages for nonresumable Pipelines #363

Show file tree
Hide file tree
Changes from 8 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
ebb5d76
[JENKINS-55287] Improve error messages for nonresumable Pipelines
dwnusbaum Jun 18, 2020
b360ae0
Jenkins.isTerminating() cannot be used from terminators
dwnusbaum Jun 19, 2020
f631a44
Merge branch 'JENKINS-55287-logging' into JENKINS-55287
dwnusbaum Jun 19, 2020
686a6e4
Clarify log messages in CpsFlowExecution.checkpoint when the build is…
dwnusbaum Jun 19, 2020
eb08b5b
Don't set persistedClean to true in CpsFlowExecution.saveOwner at all
dwnusbaum Jun 19, 2020
2214c42
Always modify persistedClean in CpsFlowExecution.checkpoint, null it …
dwnusbaum Jun 19, 2020
b3a6847
Ignore RestartingLoadStepTest.updatedBindingsOnRestart because it is …
dwnusbaum Jun 29, 2020
2ad7e70
wip
dwnusbaum Jul 2, 2020
d0c638c
Merge branch 'master' into JENKINS-55287
dwnusbaum Aug 11, 2020
22e7086
Unignore RestartingLoadStepTest.updatedBindingsOnRestart
dwnusbaum Aug 11, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
97 changes: 66 additions & 31 deletions src/main/java/org/jenkinsci/plugins/workflow/cps/CpsFlowExecution.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -712,17 +712,26 @@ public void onLoad(FlowExecutionOwner owner) throws IOException {
this.owner = owner;

try {
initializeStorage(); // Throws exception and bombs out if we can't load FlowNodes
} catch (Exception ex) {
try {
initializeStorage(); // Throws exception and bombs out if we can't load FlowNodes
} catch (Exception ex) {
LOGGER.log(Level.WARNING, "Error initializing storage and loading nodes, will try to create placeholders for: "+this, ex);
if (!canResume()) {
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Alternatively, we could check this even before we attempt to initialize storage and just bail out right away. I don't know if that is too pessimistic, maybe there is a case where the Pipeline is not resumable but flow nodes are all up to date and so the optimistic behavior here is useful because you will still be able to see the existing flow graph in those cases?

// This case is expected for PERFORMANCE_OPTIMIZED Pipelines after a Jenkins crash (JENKINS-55287). The specific issue is
// usually that only the FlowStartNode has been persisted and so we cannot find the current head node in storage.
String message = "Unable to resume " + owner.getExecutable() + " because it was not saved before Jenkins shut down. Did Jenkins crash?";
LOGGER.log(Level.WARNING, message, ex);
ex = new AbortException(message); // Change the exception displayed in the build log to one that is more useful for users.
} else {
LOGGER.log(Level.WARNING, "Unable to resume " + owner.getExecutable() + " because its flow nodes could not be loaded", ex);
}
createPlaceholderNodes(ex);
return;
} catch (Exception ex2) {
Copy link
Member Author

@dwnusbaum dwnusbaum Jun 18, 2020 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is all mostly refactoring. The old code had this structure:

try {
  try {
    // code that can throw
  } catch (first) {
    // code that can throw
    return
  }
} catch (second) {
   throw second // first exception, if any, is totally lost
}

Now it has this structure:

try {
  // code that can throw
} catch (first) {
  try {
    // code that can throw
  } catch (second) {
    second.addSuppressed(first)
    throw second
  }
  return
}

ex2.addSuppressed(ex);
done = true;
programPromise = Futures.immediateFailedFuture(ex2);
throw new IOException("Failed to even create placeholder nodes for execution", ex2);
}
} catch (Exception ex) {
done = true;
programPromise = Futures.immediateFailedFuture(ex);
throw new IOException("Failed to even create placeholder nodes for execution", ex);
return;
}

try {
Expand All @@ -737,12 +746,16 @@ public void onLoad(FlowExecutionOwner owner) throws IOException {
}
} else { // See if we can/should resume build
if (canResume()) {
// Before we resume, we need to unset persistedClean in case Jenkins restarts again.
persistedClean = null;
saveOwner();
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should add a test case for this. The test would use a PERFORMANCE_OPTIMIZED Pipeline, restart Jenkins normally, then restart Jenkins again with a hard shutdown and make sure we get the error message that the Pipeline can't be resumed rather than a raw exception.

Copy link
Member Author

@dwnusbaum dwnusbaum Jul 1, 2020 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I came up with a test case, but it doesn't fail without this line because of a random save that happens when CpsFlowExecution.notifyListeners is called. IIUC, #234 was trying to stop that save from occurring for PERFORMANCE_OPTIMIZED Pipelines, so I will look into making related changes either in this PR or before this PR is merged.

loadProgramAsync(getProgramDataFile());
} else {
// TODO if possible, consider trying to close out unterminated blocks to keep existing graph history
// That way we can visualize the graph in some error cases.
LOGGER.log(Level.WARNING, "Pipeline state not properly persisted, cannot resume "+owner.getUrl());
throw new IOException("Cannot resume build -- was not cleanly saved when Jenkins shut down.");
String message = "Unable to resume " + owner.getExecutable() + " because it was not saved before Jenkins shut down. Did Jenkins crash?";
LOGGER.log(Level.WARNING, message);
throw new IOException(message);
}
}
} catch (Exception e) { // Broad catch ensures that failure to load do NOT nuke the master
Expand Down Expand Up @@ -1526,10 +1539,6 @@ public boolean isPaused() {
return false;
}

private void setPersistedClean(boolean persistedClean) { // Workaround for some issues with anonymous classes.
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Unused

this.persistedClean = persistedClean;
}

/**
* Pause or unpause the execution.
*
Expand All @@ -1548,8 +1557,12 @@ public void pause(final boolean v) throws IOException {
if (v) {
g.pause();
checkAndAbortNonresumableBuild(); // TODO Verify if we can rely on just killing paused builds at shutdown via checkAndAbortNonresumableBuild()
checkpoint();
checkpoint(false);
} else {
// Pausing the build sets persistedClean to true so the build can resume, so if we unpause the build
// we need to unset persistedClean again.
persistedClean = null;
saveOwner();
Comment on lines +1566 to +1569
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We should add a test case for this. The Pipeline would need to be paused, then unpaused, and then we do a hard shutdown and make sure we get the error message that the Pipeline can't be resumed rather than a raw exception.

g.unpause();
}
try {
Expand Down Expand Up @@ -1591,7 +1604,7 @@ public void pause(final boolean v) throws IOException {
LOGGER.log(Level.WARNING, "Error waiting for Pipeline to suspend: "+exec, ex);
}
}
cpsExec.checkpoint();
cpsExec.checkpoint(true);
}
} catch (Exception ex) {
LOGGER.log(Level.WARNING, "Error persisting Pipeline execution at shutdown: "+((CpsFlowExecution) execution).owner, ex);
Expand Down Expand Up @@ -1954,7 +1967,6 @@ void saveOwner() {
try {
if (this.owner != null && this.owner.getExecutable() instanceof Saveable) { // Null-check covers some anomalous cases we've seen
Saveable saveable = (Saveable)(this.owner.getExecutable());
persistedClean = true;
if (storage != null && storage.delegate != null) {
// Defensively flush FlowNodes to storage
try {
Expand All @@ -1972,72 +1984,95 @@ void saveOwner() {
}
}

/** Save everything we can to disk - program, run, flownodes. */
private void checkpoint() {
/**
* Save everything we can to disk - program, run, flownodes.
* @param shuttingDown True if this checkpoint is happening because Jenkins is shutting down, false if it is happening because execution was paused.
*/
private void checkpoint(boolean shuttingDown) {
if (isComplete() || this.getDurabilityHint().isPersistWithEveryStep()) {
// Nothing to persist OR we've already persisted it along the way.
return;
}
LOGGER.log(Level.INFO, "Attempting to checkpoint all data for {0}{1}", new Object[] {
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The logging-related changes were originally part of #354 (there are some slight modifications here), so it might be easier to review that PR first.

this, (shuttingDown ? " before shutdown" : "")
});
boolean persistOk = true;
FlowNodeStorage storage = getStorage();
if (storage != null) {
try { // Node storage must be flushed first so program can be restored
storage.flush();
} catch (IOException ioe) {
persistOk=false;
LOGGER.log(Level.WARNING, "Error persisting FlowNode storage before shutdown", ioe);
LOGGER.log(Level.WARNING, "Error persisting FlowNode storage for: " + this, ioe);
}

// Try to ensure we've saved the appropriate things -- the program is the last stumbling block.
try {
final SettableFuture<Void> myOutcome = SettableFuture.create();
LOGGER.log(Level.INFO, "About to try to checkpoint the program for build"+this);
LOGGER.log(Level.FINE, "About to try to checkpoint the program for: {0}", this);
if (programPromise != null && programPromise.isDone()) {
runInCpsVmThread(new FutureCallback<CpsThreadGroup>() {
@Override
public void onSuccess(CpsThreadGroup result) {
try {
LOGGER.log(Level.INFO, "Trying to save program before shutdown "+this);
LOGGER.log(Level.FINE, "Trying to save program for: {0}", CpsFlowExecution.this);
result.saveProgramIfPossible(true);
LOGGER.log(Level.INFO, "Finished saving program before shutdown "+this);
LOGGER.log(Level.FINE, "Finished saving program for: {0}", CpsFlowExecution.this);
myOutcome.set(null);
} catch (Exception ex) {
LOGGER.log(Level.WARNING, "Error persisting program: "+ex);
// Logged at Level.WARNING when we call `myOutcome.get` and it throws an exception.
myOutcome.setException(ex);
}
}

@Override
public void onFailure(Throwable t) {
LOGGER.log(Level.WARNING, "Failed trying to save program before shutdown "+this);
// Logged at Level.WARNING when we call `myOutcome.get` and it throws an exception.
myOutcome.setException(t);
}
});
myOutcome.get(30, TimeUnit.SECONDS);
LOGGER.log(Level.FINE, "Successfully saved program for "+this);
LOGGER.log(Level.FINE, "Successfully saved program for: {0}", this);
} else {
persistOk = false;
LOGGER.log(Level.WARNING, "Unable to persist program because it was never loaded for: {0}", this);
}

} catch (TimeoutException te) {
persistOk = false;
LOGGER.log(Level.WARNING, "Timeout persisting program at execution checkpoint", te);
LOGGER.log(Level.WARNING, "Timeout persisting program for: " + this, te);
} catch (ExecutionException | InterruptedException ex) {
persistOk = false;
LOGGER.log(Level.FINE, "Error saving program, that should be handled elsewhere.", ex);
LOGGER.log(Level.WARNING, "Error saving program for: " + this, ex);
}
try { // Flush node storage just in case the Program mutated it, just to be sure
storage.flush();
LOGGER.log(Level.FINE, "Successfully did final flush of storage for "+this);
LOGGER.log(Level.FINE, "Successfully did final flush of storage for: {0}", this);
} catch (IOException ioe) {
persistOk=false;
LOGGER.log(Level.WARNING, "Error persisting FlowNode storage before shutdown", ioe);
LOGGER.log(Level.WARNING, "Error persisting FlowNode storage for: " + this, ioe);
}
persistedClean = persistOk;
try {
saveOwner();
} catch (Exception ex) {
LOGGER.log(Level.WARNING, "Error saving build for "+this, ex);
persistOk = false;
LOGGER.log(Level.WARNING, "Error saving build for: " + this, ex);
}

} else {
persistOk = false;
LOGGER.log(Level.WARNING, "No FlowNode storage for: {0}", this);
}

if (persistOk) {
LOGGER.log(Level.INFO, "Successfully checkpointed {0}{1}", new Object[] {
this, (shuttingDown ? " before shutdown" : "")
});
} else {
LOGGER.log(Level.WARNING, "Unable to successfully checkpoint {0}{1}", new Object[] {
this, (shuttingDown ? " before shutdown, so this build will probably fail when Jenkins restarts" : "")
});
}
}

Expand Down
7 changes: 3 additions & 4 deletions src/test/java/org/jenkinsci/plugins/workflow/cps/PersistenceProblemsTest.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,6 @@
import org.jenkinsci.plugins.workflow.support.steps.input.InputStepExecution;
import org.junit.Assert;
import org.junit.ClassRule;
import org.junit.Ignore;
import org.junit.Rule;
import org.junit.Test;
import org.jvnet.hudson.test.BuildWatcher;
Expand Down Expand Up @@ -244,7 +243,6 @@ public void inProgressNormal() throws Exception {
}

@Test
@Ignore
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

These tests have been ignored since they were added. I am not sure why, but I went ahead and unignored them.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You probably already knew this, but I just noticed that these test cases seem to be duplicated in workflow-job-plugin's CpsPersistenceTest. Once the dust settles, it might be worth doing a sweep of the two copies to ensure that they are in sync with each other.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah, I think I will probably delete the copies in workflow-job after this. Ideally we would run the PCT in the CI builds here with just the core Pipeline plugins so that changes are tested against the newest versions of everything.

public void inProgressMaxPerfCleanShutdown() throws Exception {
final int[] build = new int[1];
story.then( j -> {
Expand All @@ -263,14 +261,14 @@ public void inProgressMaxPerfCleanShutdown() throws Exception {
});
}

@Issue("JENKINS-55287")
@Test
@Ignore
public void inProgressMaxPerfDirtyShutdown() throws Exception {
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is the exact situation reported in JENKINS-55287 as far as I can tell.

final int[] build = new int[1];
final String[] finalNodeId = new String[1];
story.thenWithHardShutdown( j -> {
runBasicPauseOnInput(j, DEFAULT_JOBNAME, build, FlowDurabilityHint.PERFORMANCE_OPTIMIZED);
// SHOULD still save at end via persist-at-shutdown hooks
// Will not be saved since the persist-at-shutdown hooks will occur after the snapshot for thenWithHardShutdown is taken.
});
story.then( j->{
WorkflowJob r = j.jenkins.getItemByFullName(DEFAULT_JOBNAME, WorkflowJob.class);
Expand All @@ -279,6 +277,7 @@ public void inProgressMaxPerfDirtyShutdown() throws Exception {
j.waitForCompletion(run);
assertCompletedCleanly(run);
Assert.assertEquals(Result.FAILURE, run.getResult());
j.assertLogContains("was not saved before Jenkins shut down", run);
finalNodeId[0] = run.getExecution().getCurrentHeads().get(0).getId();
});
story.then(j-> {
Expand Down