cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Aleksey Yeschenko (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-5179) Hinted handoff sends over 1000 rows for one column change
Date Tue, 12 Mar 2013 04:15:13 GMT


Aleksey Yeschenko commented on CASSANDRA-5179:

4881221363f984ab6610756cab38e1a016b79e15 (CASSANDRA-4761) broke it. Current pagination logic
doesn't deal well with the fact that deleteHint is now being called from a response handler
callback and can go through the same hint again and again and again until it's finally replaced
with a tombstone. This becomes really visible in multi-dc setups, where it can take a while
to complete the write, but I was able to trigger it with ccm as well (the same hint got sent
up to 3 times).

Should we reopen CASSANDRA-4761 or deal with it here?
> Hinted handoff sends over 1000 rows for one column change
> ---------------------------------------------------------
>                 Key: CASSANDRA-5179
>                 URL:
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 1.2.2
>         Environment: Windows 7
> Java 1.6u38
>            Reporter: Antti Koivisto
>            Assignee: Aleksey Yeschenko
>             Fix For: 1.2.3
>         Attachments: cassandra_receiver.log, cassandra_sender.log
> We have a small test environment with two datacenters (DC1 and DC2) running on Windows
7 laptops.
> Both datacenters have one node. We use network topology strategy to replicate all data
to both datacenters.
> We started with empty db. 
> 1. Created a keyspace with strategy options [DC1:1, DC2:1]
> 2. Added one row to a column family with CLI to DC1. Change was replicated to DC2.
> 3. Disconnected network cable from DC2.
> 4. Gossiper noticed, that other DC is dead.
> 5. Added another row to DC1.
> 6. Reconnected cable on DC2.
> 7. DC1 started hinted handoff for DC2.
> 8. Hinted handoff is finished with message: "Finished hinted handoff of 1969 rows to
endpoint <DC2 ip>"
> We repeated test with same results on Linux cluster with Cassandra 1.2.0. 
> On Cassandra 1.1.5 Linux cluster, only one row was sent to endpoint. "Finished hinted
handoff of 1 rows to endpoint <DC2 ip>"

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see:

View raw message