Adar Dembo has posted comments on this change.

Change subject: master: do not delete unknown tablets

Patch Set 5:

Commit Message:

Line 7: master: do not delete unknown tablets
> I wonder if we now need some tool to "reformat" a tablet server? or to inse
I'm almost positive that a user will one day blow away their master metadata 
directory and ask us to fix it. IIRC this happened with HDFS in the CDH3 days. 
The aggregate of all  tablet metadata should be sufficient to reconstruct the 
master metadata, provided all of the tablets are sufficiently replicated. But, 
I don't know if it makes sense to spend any time on this problem now, given 
that no one has run into this and our other priorities.

Reformatting a tserver is pretty easy: just stop it, delete its wal/data 
directories, and restart it. Or did you mean something else?

I don't really see a use case for dummy "deleted table/tablet" records, apart 
from the aforementioned recovery case, at which point these aren't "dummy" 
records. But, deleting orphaned tablets to reclaim space is a legit use case. 
The minimal fix is to add a gflag that defaults to off and grants the master 
permission to delete orphaned tablets. I'll do that here.

Beyond that, let me know what JIRAs I should file. :)

Line 9: Quoting from the multi-master design doc:
> nit: can you provide a link or source code path here?
File src/kudu/master/

PS5, Line 1546: INFO)
> WARNING might be more appropriate here? or perhaps INFO if it's an incremen
I changed it to WARNING. It's hard to know in this context whether it's an 
incremental report or not, and besides, an incremental report only includes a 
tablet if its state has changed in some way.

To view, visit
To unsubscribe, visit

Gerrit-MessageType: comment
Gerrit-Change-Id: I331f2d5bb06c38daa7b09854dbb24a7881723551
Gerrit-PatchSet: 5
Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-Owner: Adar Dembo <>
Gerrit-Reviewer: Adar Dembo <>
Gerrit-Reviewer: Dan Burkert <>
Gerrit-Reviewer: Mike Percy <>
Gerrit-Reviewer: Todd Lipcon <>
Gerrit-HasComments: Yes

Reply via email to