[
https://issues.apache.org/jira/browse/CASSANDRA-13115?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Xiaolong Jiang updated CASSANDRA-13115:
---------------------------------------
Description:
The code trying to wait(block) for repair result to come back in 3.X is below:
bq. public void close()
{
try
{
FBUtilities.waitOnFutures(repairResults,
DatabaseDescriptor.getWriteRpcTimeout());
}
catch (TimeoutException ex)
{
// We got all responses, but timed out while repairing
int blockFor = consistency.blockFor(keyspace);
if (Tracing.isTracing())
Tracing.trace("Timed out while read-repairing after
receiving all {} data and digest responses", blockFor);
else
logger.debug("Timeout while read-repairing after receiving
all {} data and digest responses", blockFor);
throw new ReadTimeoutException(consistency, blockFor-1,
blockFor, true);
}
}
in DataResolver class, but this close method is never called and it's also not
auto close(RepairMergeListener is not extending from
AutoCloseable/CloseableIterator) which means we never wait for repair to finish
before returning final result.
The steps to reproduce:
1. create some keyspace/table with RF = 2
2. start 2 nodes using ccm
3. stop node2
4. disable node1 hinted hand off
5. write some data to node1 with consistency level one
6. start node2
7. query some data from node1
This should trigger read repair. I put some log in above close method, and can
not see log print put.
So this bug will basically violate "monotonic quorum reads " guarantee.
was:
The code trying to wait(block) for repair result to come back in 3.X is below:
{code}
public void close()
{
try
{
FBUtilities.waitOnFutures(repairResults,
DatabaseDescriptor.getWriteRpcTimeout());
}
catch (TimeoutException ex)
{
// We got all responses, but timed out while repairing
int blockFor = consistency.blockFor(keyspace);
if (Tracing.isTracing())
Tracing.trace("Timed out while read-repairing after
receiving all {} data and digest responses", blockFor);
else
logger.debug("Timeout while read-repairing after receiving
all {} data and digest responses", blockFor);
throw new ReadTimeoutException(consistency, blockFor-1,
blockFor, true);
}
}
{code}
in DataResolver class, but this close method is never called and it's also not
auto close(RepairMergeListener is not extending from
AutoCloseable/CloseableIterator) which means we never wait for repair to finish
before returning final result.
The steps to reproduce:
1. create some keyspace/table with RF = 2
2. start 2 nodes using ccm
3. stop node2
4. disable node1 hinted hand off
5. write some data to node1 with consistency level one
6. start node2
7. query some data from node1
This should trigger read repair. I put some log in above close method, and can
not see log print put.
So this bug will basically violate "monotonic quorum reads " guarantee.
> Read repair is not blocking repair to finish in foreground repair
> -----------------------------------------------------------------
>
> Key: CASSANDRA-13115
> URL: https://issues.apache.org/jira/browse/CASSANDRA-13115
> Project: Cassandra
> Issue Type: Bug
> Environment: ccm on OSX
> Reporter: Xiaolong Jiang
>
> The code trying to wait(block) for repair result to come back in 3.X is below:
> bq. public void close()
> {
> try
> {
> FBUtilities.waitOnFutures(repairResults,
> DatabaseDescriptor.getWriteRpcTimeout());
> }
> catch (TimeoutException ex)
> {
> // We got all responses, but timed out while repairing
> int blockFor = consistency.blockFor(keyspace);
> if (Tracing.isTracing())
> Tracing.trace("Timed out while read-repairing after
> receiving all {} data and digest responses", blockFor);
> else
> logger.debug("Timeout while read-repairing after
> receiving all {} data and digest responses", blockFor);
> throw new ReadTimeoutException(consistency, blockFor-1,
> blockFor, true);
> }
> }
> in DataResolver class, but this close method is never called and it's also
> not auto close(RepairMergeListener is not extending from
> AutoCloseable/CloseableIterator) which means we never wait for repair to
> finish before returning final result.
> The steps to reproduce:
> 1. create some keyspace/table with RF = 2
> 2. start 2 nodes using ccm
> 3. stop node2
> 4. disable node1 hinted hand off
> 5. write some data to node1 with consistency level one
> 6. start node2
> 7. query some data from node1
> This should trigger read repair. I put some log in above close method, and
> can not see log print put.
> So this bug will basically violate "monotonic quorum reads " guarantee.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
