[
https://issues.apache.org/jira/browse/KAFKA-188?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jay Kreps updated KAFKA-188:
----------------------------
Attachment: KAFKA-188.patch
Okay, took a shot at this. Attached is a draft patch. It needs some tests
specific to the log selection.
There is a hitch, which I discuss below.
The basic change is
1. Change KafkaConfig.logDir to logDirectories and take a CSV of paths (I still
support the old setting as a fallback)
2. Move the whole clean shutdown marker file thing into the LogManager. I feel
it should have always been inside LogManager since it is an implementation
detail of log recovery. Now there is one such file per log directory.
3. Create new logs in the directories in a round-robin fashion. To do this I
keep a counter in LogManager that controls the index of the dir in which we
will next create a new log. Initialize this to a random dir index. Each time we
create a new index use the dir this points at and then increment it.
The hitch is the high watermark file. Currently we keep it in log.dir. But what
directory should we keep it in when there are multiple log directories? I
hackily just use the first. However this creates a dependency on the order of
the log dirs in the config, which is not ideal. If we sort them and use the dir
that is alphabetically first then if we add new directories that will mess it
up (whereas if we hadn't sorted we could have tacked them on the end).
Some options:
0. What I currently do, just use the first directory for the hwmark file.
1. Add a new configuration property, metadataDir and use this for the
highwatermark file and the clean shutdown marker and any future persistent
thing. Downside of this is that it requires a new config parameter the user has
to set up.
2. Require a top level directory for all log directories. e.g.
all_logs/
log_dir_1/
my_topic-1/
my_topic-2
log_dir_2/
...
Obviously since the goal is to support multiple independently mounted disks and
you might not control where they mount to you might have to use soft links.
From a previous life I am remembering lots of pain related to java and
softlinks.
I would like to get people's feedback on this.
> Support multiple data directories
> ---------------------------------
>
> Key: KAFKA-188
> URL: https://issues.apache.org/jira/browse/KAFKA-188
> Project: Kafka
> Issue Type: New Feature
> Reporter: Jay Kreps
> Attachments: KAFKA-188.patch
>
>
> Currently we allow only a single data directory. This means that a multi-disk
> configuration needs to be a RAID array or LVM volume or something like that
> to be mounted as a single directory.
> For a high-throughput low-reliability configuration this would mean RAID0
> striping. Common wisdom in Hadoop land has it that a JBOD setup that just
> mounts each disk as a separate directory and does application-level balancing
> over these results in about 30% write-improvement. For example see this claim
> here:
> http://old.nabble.com/Re%3A-RAID-vs.-JBOD-p21466110.html
> It is not clear to me why this would be the case--it seems the RAID
> controller should be able to balance writes as well as the application so it
> may depend on the details of the setup.
> Nonetheless this would be really easy to implement, all you need to do is add
> multiple data directories and balance partition creation over these disks.
> One problem this might cause is if a particular topic is much larger than the
> others it might unbalance the load across the disks. The partition->disk
> assignment policy should probably attempt to evenly spread each topic to
> avoid this, rather than just trying keep the number of partitions balanced
> between disks.
--
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: http://www.atlassian.com/software/jira
