Adar Dembo has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/12239 )

Change subject: [fs]: delete the (orphaned) metadata file when repairing
......................................................................


Patch Set 8:

(1 comment)

http://gerrit.cloudera.org:8080/#/c/12239/6/src/kudu/fs/log_block_manager.cc
File src/kudu/fs/log_block_manager.cc:

http://gerrit.cloudera.org:8080/#/c/12239/6/src/kudu/fs/log_block_manager.cc@772
PS6, Line 772: con
> Won't this case be handled by the existing "delete dead containers at start
Hao and I discussed this on Slack. She pointed to the corresponding test case, 
where the LBM starts up with Status::Corruption. That's because the data file 
size was really 0, which meant that the dead blocks described the metadata file 
were treated as malformed in ProcessRecord(). I overlooked that, thinking that 
while the _number of bytes on disk_ was 0, the file size was something closer 
to FLAGS_log_container_max_size.

We could address that by deferring block consistency checks until after all 
DELETEs are processed, thus only checking live blocks. Later, the dead 
container cleanup code would remove the container.

But I don't think that needs to be addressed here, because the goal of this 
patch is to clean up the so-called "half present" containers, and in that case, 
the data file is missing outright. I'm not sure under what circumstances 
someone would end up with a zero-length data file; perhaps they accidentally 
truncated the data file? Or ran with fsync disabled and created+deleted one 
block?



--
To view, visit http://gerrit.cloudera.org:8080/12239
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I2a4223d0e6385a7f7776f8fe2679a3bb16d832ab
Gerrit-Change-Number: 12239
Gerrit-PatchSet: 8
Gerrit-Owner: helifu <[email protected]>
Gerrit-Reviewer: Adar Dembo <[email protected]>
Gerrit-Reviewer: Hao Hao <[email protected]>
Gerrit-Reviewer: Kudu Jenkins (120)
Gerrit-Reviewer: helifu <[email protected]>
Gerrit-Comment-Date: Fri, 25 Jan 2019 00:08:17 +0000
Gerrit-HasComments: Yes

Reply via email to