cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sylvain Lebresne (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-11115) Thrift removal
Date Wed, 07 Dec 2016 14:01:58 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-11115?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15728839#comment-15728839
] 

Sylvain Lebresne commented on CASSANDRA-11115:
----------------------------------------------

I push commits that address I believe the points above.

bq. In LegacyLayout, treatment of the removed {{readAllAsDynamic}} arg as always false.

That was a good point, I did slightly badly translated that. That said, the {{encodeCellName}}
was only used by {{CounterCacheKey}} that was encoding everything as an old format cell name.
And as it happens, {{CounterCacheKey}} was encoding manually and wasn't bothering with compact
table complications, so I think using {{encodeCellName}} was wrong in the first place (as
in, before this patch), which we might not have noticed because the only consequence is that
the counter cache don't get pre-warmed properly. Anyway, this can be handled much more easily
and cleanly in {{CounterCacheKey}} directly and I added a commit to do that, allowing to remove
more code from {{LegacyLayout}}.

bq. {{CqlInputFormat}}/{{ConfigHelper}} whether or not nulls where valid/legal before for
job range start/end keys/tokens

The previous code was assuming that if {{start_token}} wasn't {{null}}, then {{end_token}}
wasn't (it was using {{partitioner.getTokenFactory().fromString()}} which doesn't support
nulls). So the previous code was allowing {{setInputRange(null, null)}} and {{setInputRange(null,
<something>)}}, both of which were doing nothing, but *not* {{setInputRange(<something>,
null)}}. Which to me suggests nobody could rely on this on purpose and not supporting nulls
at all is fine. Happy to strictly preserve the previous behavior though if you're worried.


An update on the dtests: I've been able to run them with the modified CCM branch from above,
and they need a few updates to skip a few thrift tests and stop using removed yaml properties.
I have a branch for this [here|https://github.com/pcmanus/cassandra-dtest/commits/thrift_removal_tests].
The last run that should hopefully be clean or close to it is ongoing:
| [11115|https://github.com/pcmanus/cassandra/commits/11115] | [utests|http://cassci.datastax.com/job/pcmanus-11115-testall]
| [dtests|http://cassci.datastax.com/job/pcmanus-11115-dtest] |


> Thrift removal
> --------------
>
>                 Key: CASSANDRA-11115
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-11115
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Sylvain Lebresne
>            Assignee: Sylvain Lebresne
>             Fix For: 4.0
>
>
> Thrift removal [has been announced for 4.0|http://mail-archives.apache.org/mod_mbox/cassandra-user/201601.mbox/%3CCALdd-zgAgnLDu3pQBd6wp0jB0X73QjDr9PhPXMMO+gQ+2E5WcQ@mail.gmail.com%3E].
This ticket is meant to serve as a general task for that removal, but also to track issue
related to that, either things that we should do in 3.x to make that removal as smooth as
possible, or sub-tasks that it makes sense to separate.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message