hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Erik Holstad (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HBASE-1249) Rearchitecting of server, client, API, key format, etc for 0.20
Date Sun, 29 Mar 2009 18:18:50 GMT

    [ https://issues.apache.org/jira/browse/HBASE-1249?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12693596#action_12693596

Erik Holstad commented on HBASE-1249:

When bringing DeleteFamily into the mix it creates problems with the current layout, cause
every time you need to merge the deletes from the previous storefile with the current deletes
and every time you compare those deletes with the current position that you are looking at
in the current storefile you need to compare those timestamps. There are a couple of ways
around this as I see it.

1. Only letting the user set the timestamp for the DeleteFamily to now, where now can be System.currentTimeMillis()
or some user generated now. This would mean the current memCache would be cleaned and all
other storefiles considered to be deleted, for this row and family. This would mean that you
only need to do one check for every storefile to see if there is a DeleteFamily entry in there
and you will know that all data in that storefile is ok and that you don't need to look in
any more storefiles.

2. Do the check if there is a DeleteFamily in the deletes and have 2 different methods taking
care of the cases where you have a DeleteFamily entry and when you don't. The downside of
this is that you still have to pay the cost if you do have a DeleteFamily, worst case you
have to do these 2 checks for every entry in every storefile.

3. Keep the DeleteFamily sorted at the timestamp where it belongs, so that all deletes would
be sorted in timestamp order before column. This is a rather big change, because it means
that also the puts would have to be sorted this way for it to make any sense. Another advantage
of this approach would be that earlying out from a query with a timerange "filter" would be
more effective.

I personally like the 3 option the most, but I can see people not liking it because it sort
of redefines what HBase is, so I think that number 1 is the best option and after that number

Would love to get some input in this matter to see if there is anything that people might
have against number 1. Otherwise I will move on with the process of implementing that option.

Regards Erik

> Rearchitecting of server, client, API, key format, etc for 0.20
> ---------------------------------------------------------------
>                 Key: HBASE-1249
>                 URL: https://issues.apache.org/jira/browse/HBASE-1249
>             Project: Hadoop HBase
>          Issue Type: Improvement
>            Reporter: Jonathan Gray
>            Priority: Blocker
>             Fix For: 0.20.0
>         Attachments: HBASE-1249-Example-v1.pdf, HBASE-1249-Example-v2.pdf, HBASE-1249-GetQuery-v1.pdf,
HBASE-1249-GetQuery-v2.pdf, HBASE-1249-GetQuery-v3.pdf, HBASE-1249-StoreFile-v1.pdf
> To discuss all the new and potential issues coming out of the change in key format (HBASE-1234):
zero-copy reads, client binary protocol, update of API (HBASE-880), server optimizations,

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message