Adar Dembo has posted comments on this change.

Change subject: log block manager: use extent maps to decide whether to 
truncate containers
......................................................................


Patch Set 6:

> Yea, I think it's quite reasonable to only do so in a "fs check
 > --repair" type situation.
 
OK, I'll work on that.

 > Another option I think is reasonable is to use stat() on the
 > container file, see the actual space usage, and if it's
 > significantly more than we expect based on the live block count in
 > the container, trigger the hole-punching.
 
Right, we talked about this before. Assuming an "expensive repair" thing 
exists, the issue isn't knowing _when_ to re-hole-punch; it's knowing _how 
much_ unpunched space exists (for reporting purposes), and this approach 
doesn't provide an accurate measure of that.

 > A third option: when we successfully hole-punch after a delete,
 > write an entry to the metadata file saying "hole-punched!" (no need
 > to fsync it). On restart, re-punch any containers which have
 > deletions at the trail of the metadata with no "hole-punched!".
 > This way you'd only have to re-punch those files which might have
 > missed their chance due to a premature process exit last time
 > around.

That's an interesting idea, though it trades off extent map fetching for larger 
metadata files. Need to think more about it.

 > BTW the restart times are longer than i expected, would be curious
 > to know where all that time is going (but that's unrelated to this
 > jira)

I'll look into that later.

-- 
To view, visit http://gerrit.cloudera.org:8080/6585
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I3541f70721328d5af911f4889f8b4bcf0de3ebf8
Gerrit-PatchSet: 6
Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-Owner: Adar Dembo <a...@cloudera.com>
Gerrit-Reviewer: Adar Dembo <a...@cloudera.com>
Gerrit-Reviewer: David Ribeiro Alves <dral...@apache.org>
Gerrit-Reviewer: Kudu Jenkins
Gerrit-Reviewer: Todd Lipcon <t...@apache.org>
Gerrit-HasComments: No

Reply via email to