Details
-
Bug
-
Status: Resolved
-
Minor
-
Resolution: Won't Fix
-
1.6.6, 1.7.2
-
None
-
None
Description
Saw this on 1.6.6 with a small disk for testing (32GB disk). The default walog size is around 1GB, and only 3.4GB were left available on each data node.
The namenode reported that no data nodes had space available when trying to write the first block, so the tserver failed to write the file. It kept retrying, resulting in the namenode filling up with thousands of zero-length WAL files.
The fix was to lower the tserver.walog.max.size to 100M. Another solution would be to use a larger disk.
The infinite loop problem, constantly creating new empty WAL files is still a problem, but it should only happen when low on disk space, which is likely going to cause other, more serious problems... and could be avoided with good system monitoring.
I have not tested on versions newer than 1.6.6, but I imagine it's still a problem.
Attachments
Issue Links
- links to