Pnfs Release Notes 3.1.9

Compatibility

Bug Fixes

File name length

File names and link names may now have up to 199 characters without the special pnfs characters ".(...". This is not true for directory names. Directory names are limited to 199 characters, pnfs specials included.

Removing level files

When removing special pnfs level files, the file itself wasn't changed but the contents of the different levels were written into the trash directories as if the master (level 0) had been removed. This caused the backend HSM to remove the file from tape without changing the pnfs filesystem entry. (very nasty problem).
Has been fixed.

Behaviour on high load

There are two new compile time flags defining the behaviour of the pnfs deamons if the number of requests exceeds the maximum performance of the database servers.

POST_TIMEOUT=<#ofSeconds>

This options defines the number of seconds the pnfs server will wait for the database server to become available.

NO_ANSWER_ON_TIMEOUT

This option defines the type of answer, the pnfs deamon replies to the client in case the database server doesn't become available within POST_TIMEOUT. If NO_ANSWER_ON_TIMEOUT is defined, the pnfs server won't answer at all, otherwise it sends an NO SUCH FILE OR DIRECTORY.

In case the pnfs server is hard mounted, the POST_TIMEOUT should be set to the nfs client retry time and NO_ANSWER_ON_TIMEOUT should be defined.

Modifications for dCache compatibility

Some pnfs special commands modify file attributes like filesize or uid/gid by pnfsId after the object has been created. Unfortunately the solaris nfs clients send a nfs command setting the group id after doing the required operation to the id of the sender of the command, which in terms of the dCache is wrong because the cache always is running as root. The oberved effect was that after successfully creating a file with the correct uid/gid of the user, the file suddenly changed to the gid to 1 or 0 as soon as the dCache has set the correct filesize. Something similiar happens if the dCache has to create the file entry. It will do so as root and later on uses a special pnfs command to change the uid/gid. While the uid always changed sucessfully the gid never actually changed if the dCache PnfsManager has been running on solaris machines.
Has been fixed.

Security preparation