We deeply regret to inform GPFS users that the 220.127.116.11 service update contains a serious flaw. When parsing some RPCs sent by nodes running GPFS 18.104.22.168 or earlier, the code in 22.214.171.124 incorrectly processes certain data structures, resulting in GPFS crashes. The most typical symptom of this issue is the "((diskNum > 0) && (diskNum <= sgP->getNDisks())" assert in LogFile::append, although other symptoms are possible. The problem occurs most readily when metadata is not replicated (mmlsfs shows -m 1); metadata replication mitigates the problem to a large extent, but not completely. Full data and metadata replication would prevent the problem from happening.
The problem is limited to the scenarios where 126.96.36.199 and pre-188.8.131.52 nodes are present in the same cluster, accessing the same file systems. If you are considering upgrading GPFS from a pre-184.108.40.206 level (including 3.3.0.X), do not upgrade to the unpatched 220.127.116.11. Clusters running only 18.104.22.168 or later nodes, including 22.214.171.124, are not affected.
The problem will be fixed in 126.96.36.199. An efix for 188.8.131.52 is also available via IBM Service.
This topic has been locked.
1 reply Latest Post - 2012-05-02T18:48:32Z by firstname.lastname@example.org
Pinned topic Backward compatibility problem in GPFS 184.108.40.206
Answered question This question has been answered.
Unanswered question This question has not been answered yet.
Updated on 2012-05-02T18:48:32Z at 2012-05-02T18:48:32Z by email@example.com
firstname.lastname@example.org 060000T9GF202 PostsACCEPTED ANSWER
Re: Backward compatibility problem in GPFS 220.127.116.11012-05-02T18:48:32Z in response to email@example.comTo expand on the previous post: both 18.104.22.168 and 22.214.171.124 levels contain the same backward compatibility issue. All of the 126.96.36.199 warnings and workarounds also apply to 188.8.131.52.