Skip to content

CASSANDRA-20730: Fix topology reporting races during startup #4212

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

Closed
ifesdjeen wants to merge 1 commit into from
+19 −14
Closed

CASSANDRA-20730: Fix topology reporting races during startup #4212

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
5 changes: 3 additions & 2 deletions src/java/org/apache/cassandra/service/accord/AccordConfigurationService.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -257,15 +257,16 @@ void reportMetadataInternal(ClusterMetadata metadata)

private void checkIfNodesRemoved(Topology topology, Set<Node.Id> stillLiveNodes)
{
if (epochs.minEpoch() == topology.epoch()) return;
long minEpoch = epochs.minEpoch();
if (minEpoch == 0 || topology.epoch() <= minEpoch) return;
Topology previous = getTopologyForEpoch(topology.epoch() - 1);
// for all nodes removed, or pending removal, mark them as removed so we don't wait on their replies
Set<Node.Id> removedNodes = Sets.difference(previous.nodes(), topology.nodes());
removedNodes = Sets.filter(removedNodes, id -> !stillLiveNodes.contains(id));
// TODO (desired, efficiency): there should be no need to notify every epoch for every removed node
for (Node.Id removedNode : removedNodes)
{
if (topology.epoch() >= epochs.minEpoch())
if (topology.epoch() >= minEpoch)
onNodeRemoved(topology.epoch(), previous, removedNode);
}
}
Expand Down
28 changes: 16 additions & 12 deletions src/java/org/apache/cassandra/service/accord/AccordService.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -382,20 +382,15 @@ public synchronized void unsafeStartupWithOverrides(@Nullable Journal.TopologyUp

if (remote != null)
remote.forEach(configService::reportTopology, highestKnown + 1, Integer.MAX_VALUE);
else if (images.isEmpty()) // First boot, single-node cluster
configService.reportTopology(AccordTopology.createAccordTopology(metadata));

// Subscribe to TCM events
ClusterMetadataService.instance().log().addListener(configService.listener);
{
metadata = ClusterMetadata.current();
highestKnown = configService.currentEpoch();
if (metadata.epoch.getEpoch() > highestKnown)
{
remote = fetchTopologies(highestKnown + 1);
if (remote != null)
remote.forEach(configService::reportTopology, highestKnown + 1, Integer.MAX_VALUE);
}
}

// We report current topology _after_ subscribing to TCM events since in a single-node cluster there
// will be no notification about the current epoch, since it's already reported. And in a multi-node cluster
// we do not want a race between addinga listener and reporting an epoch.
if (remote == null && images.isEmpty())
configService.reportTopology(AccordTopology.createAccordTopology(metadata));

WatermarkCollector.fetchAndReportWatermarksAsync(configService());
configService.unsafeMarkTruncated();
Expand All @@ -414,6 +409,15 @@ else if (images.isEmpty()) // First boot, single-node cluster
catch (TimeoutException e)
{
logger.warn("Epoch {} is not ready after waiting for {} seconds", metadata.epoch, (++attempt) * waitSeconds);
// In case there are any gaps, fetch unknown topologies.
metadata = ClusterMetadata.current();
highestKnown = configService.currentEpoch();
if (metadata.epoch.getEpoch() > highestKnown)
{
remote = fetchTopologies(highestKnown + 1);
if (remote != null)
remote.forEach(configService::reportTopology, highestKnown + 1, Integer.MAX_VALUE);
}
}

if (Clock.Global.nanoTime() > deadine)
Expand Down