[
https://issues.apache.org/jira/browse/CASSANDRA-19776?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17950979#comment-17950979
]
Branimir Lambov commented on CASSANDRA-19776:
---------------------------------------------
This phrases the problem as one of CANONICAL containing more that it should,
which is not how we should look at it.
CANONICAL sstables are the state before ongoing compactions. The live state
replaces some of these with early-open sstables, which are moved to the
canonical set when they are complete and the transaction commits. The canonical
sstables are still valid for certain uses, especially for collecting metadata,
and must be present until a new canonical set is constructed. If the
transaction aborts, the canonical set becomes the live set again.
The main suspect to me is whether transaction checkpointing can cause an
sstable to be released prematurely. [This
line|https://github.com/blambov/cassandra/blob/trunk/src/java/org/apache/cassandra/db/lifecycle/LifecycleTransaction.java#L377]
seems to make that possible. Perhaps we need to obsolete the expired sstables
at the start of the compaction so that they are always in the
{{staged.obsolete}} set?
> Spinning trying to capture readers
> ----------------------------------
>
> Key: CASSANDRA-19776
> URL: https://issues.apache.org/jira/browse/CASSANDRA-19776
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Legacy/Core
> Reporter: Cameron Zemek
> Assignee: Stefan Miklosovic
> Priority: Normal
> Fix For: 4.0.x, 4.1.x, 5.0.x, 5.x
>
> Attachments: extract.log
>
> Time Spent: 20m
> Remaining Estimate: 0h
>
> On a handful of clusters we are noticing Spin locks occurring. I traced back
> all the calls to the EstimatedPartitionCount metric (eg.
> org.apache.cassandra.metrics:type=Table,keyspace=testks,scope=testcf,name=EstimatedPartitionCount)
> Using the following patched function:
> {code:java}
> public RefViewFragment selectAndReference(Function<View,
> Iterable<SSTableReader>> filter)
> {
> long failingSince = -1L;
> boolean first = true;
> while (true)
> {
> ViewFragment view = select(filter);
> Refs<SSTableReader> refs = Refs.tryRef(view.sstables);
> if (refs != null)
> return new RefViewFragment(view.sstables, view.memtables,
> refs);
> if (failingSince <= 0)
> {
> failingSince = System.nanoTime();
> }
> else if (System.nanoTime() - failingSince >
> TimeUnit.MILLISECONDS.toNanos(100))
> {
> List<SSTableReader> released = new ArrayList<>();
> for (SSTableReader reader : view.sstables)
> if (reader.selfRef().globalCount() == 0)
> released.add(reader);
> NoSpamLogger.log(logger, NoSpamLogger.Level.WARN, 1,
> TimeUnit.SECONDS,
> "Spinning trying to capture readers {},
> released: {}, ", view.sstables, released);
> if (first)
> {
> first = false;
> try {
> throw new RuntimeException("Spinning trying to
> capture readers");
> } catch (Exception e) {
> logger.warn("Spin lock stacktrace", e);
> }
> }
> failingSince = System.nanoTime();
> }
> }
> }
> {code}
> Digging into this code I found it will fail if any of the sstables are in
> released state (ie. reader.selfRef().globalCount() == 0).
> See the extract.log for an example of one of these spin lock occurrences.
> Sometimes these spin locks last over 5 minutes. Across the worst cluster with
> this issue, I ran a log processing script that everytime the 'Spinning trying
> to capture readers' was different to previous one it would output if the
> released tables were in Compacting state. Every single occurrence has it spin
> locking with released listing a sstable that is compacting.
> In the extract.log example its spin locking saying that nb-320533-big-Data.db
> has been released. But you can see prior to it spinning that sstable is
> involved in a compaction. The compaction completes at 01:03:36 and the
> spinning stops. nb-320533-big-Data.db is deleted at 01:03:49 along with the
> other 9 sstables involved in the compaction.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
