[lustre-discuss] MDTs unavailable and Recovery failed

[Sonia Sharma]

Dear Lustre Community,

I am seeking some help in understanding the errors I am seeing in my Lustre file system deployment. I would really appreciate if someone can share more information around the errors that I am seeing.

So, I have a Lustre deployment with 15 MDSs and around 250 OSSs. They all mounted fine and my client could mount Lustre as well.
But after some time, I see that 3 of the MDTs are showing as “Unavailable” (from lfs check servers) to the Client node.
NOTE  - LNet pings from Client node to these MDSs are successful all this while.

One error message that I noted from client logs is (below) this failure to get the sequence allocated from the Unavailable MDTs with rc =-4 (Interrupted system call).
I do not see these errors repeated but I see from the logs, recovery on those MDTs starting and failing when I run “df -h” on client node.

(All the MDTs/OSTs had got the super-sequence allocated fine from MDT0 and have verifies there is no conflict on those).
[111271.621088] LustreError: 11-0: lustrefs-MDT0002-mdc-ff4443340524d800: operation mds_connect to node 10.0.200.179 at tcp failed: rc = -11
[111435.913475] LustreError: 1716549:0:(fid_request.c:233:seq_client_alloc_seq()) cli-cli-lustrefs-MDT0002-mdc-ff4443340524d800: Cannot allocate new meta-sequence: rc = -4
[111435.913482] LustreError: 1716549:0:(fid_request.c:335:seq_client_alloc_fid()) cli-cli-lustrefs-MDT0002-mdc-ff4443340524d800: Can't allocate new sequence: rc = -4
[111788.935390] LustreError: 1722378:0:(fid_request.c:233:seq_client_alloc_seq()) cli-cli-lustrefs-MDT0003-mdc-ff4443340524d800: Cannot allocate new meta-sequence: rc = -4
[111788.935396] LustreError: 1722378:0:(fid_request.c:335:seq_client_alloc_fid()) cli-cli-lustrefs-MDT0003-mdc-ff4443340524d800: Can't allocate new sequence: rc = -4

If someone can let me know how/if possible, to dump the FLDB on mdt0, that would be great as well (just for knowledge).

Best regards,
Sonia
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20250606/aaba769d/attachment.htm>