[
https://issues.apache.org/jira/browse/CASSANDRA-11349?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15227163#comment-15227163
]
Fabien Rousseau commented on CASSANDRA-11349:
---------------------------------------------
Using the RangeTombstone.Tracker can help in the situation described just above.
In fact, the RT should always update the tracker (see CASSANDRA-11477).
The trick here is to always considered it as "expired" in the tracker (even if
not) so the tombstones are not accumulated during compaction (if expired the
tracker keeps only the list of opened RTs and if not, it keeps all unwritten
RTs, ie all RTs because it's a validation compaction...).
Having a look at the update method of the Tracker, it already check if the
tombstone is superseded by another one (and don't add it as "opened" if
superseded).
Thus, the v2 patch:
- includes the previous patch
- always update the tracker with the RT (considering it as expired even if
not, just to not retain too many of them in memory, and because it's for
validation, it's a read only and won't affect anything)
- test if the RT was added in the openedTombstones list, and if that's not the
case, skip it for digest.
I know that the patch may be a bit rough (at least on the "isLastOpened"
method) but it is more to validate the approach first and did not want the
patch to be too invasive (by modifying the returned value of the update method).
WDYT ?
Note: I have not yet tested it against our production data
Note2: Regarding the read-repair, this seems to be a different story and can't
see anything for now that could explain those differences (will dig later on
this as this is less urgent)
> MerkleTree mismatch when multiple range tombstones exists for the same
> partition and interval
> ---------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-11349
> URL: https://issues.apache.org/jira/browse/CASSANDRA-11349
> Project: Cassandra
> Issue Type: Bug
> Reporter: Fabien Rousseau
> Assignee: Stefan Podkowinski
> Labels: repair
> Fix For: 2.1.x, 2.2.x
>
> Attachments: 11349-2.1.patch
>
>
> We observed that repair, for some of our clusters, streamed a lot of data and
> many partitions were "out of sync".
> Moreover, the read repair mismatch ratio is around 3% on those clusters,
> which is really high.
> After investigation, it appears that, if two range tombstones exists for a
> partition for the same range/interval, they're both included in the merkle
> tree computation.
> But, if for some reason, on another node, the two range tombstones were
> already compacted into a single range tombstone, this will result in a merkle
> tree difference.
> Currently, this is clearly bad because MerkleTree differences are dependent
> on compactions (and if a partition is deleted and created multiple times, the
> only way to ensure that repair "works correctly"/"don't overstream data" is
> to major compact before each repair... which is not really feasible).
> Below is a list of steps allowing to easily reproduce this case:
> {noformat}
> ccm create test -v 2.1.13 -n 2 -s
> ccm node1 cqlsh
> CREATE KEYSPACE test_rt WITH replication = {'class': 'SimpleStrategy',
> 'replication_factor': 2};
> USE test_rt;
> CREATE TABLE IF NOT EXISTS table1 (
> c1 text,
> c2 text,
> c3 float,
> c4 float,
> PRIMARY KEY ((c1), c2)
> );
> INSERT INTO table1 (c1, c2, c3, c4) VALUES ( 'a', 'b', 1, 2);
> DELETE FROM table1 WHERE c1 = 'a' AND c2 = 'b';
> ctrl ^d
> # now flush only one of the two nodes
> ccm node1 flush
> ccm node1 cqlsh
> USE test_rt;
> INSERT INTO table1 (c1, c2, c3, c4) VALUES ( 'a', 'b', 1, 3);
> DELETE FROM table1 WHERE c1 = 'a' AND c2 = 'b';
> ctrl ^d
> ccm node1 repair
> # now grep the log and observe that there was some inconstencies detected
> between nodes (while it shouldn't have detected any)
> ccm node1 showlog | grep "out of sync"
> {noformat}
> Consequences of this are a costly repair, accumulating many small SSTables
> (up to thousands for a rather short period of time when using VNodes, the
> time for compaction to absorb those small files), but also an increased size
> on disk.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
