(Ended) (Follow-up)[Outage] June 18, 2024: Emergency maintenance of Lustre7
The restoration work was completed at 9:00am on Tuesday, 25 June 2024.
Also, the gateways (gw.ddbj.nig.ac.jp, gw2.ddbj.nig.ac.jp) have been restored and are available for logging in.
- At 18:21:14 on Mon 17 Jun, a fault occurred on the Lustre7 high-speed storage system in the General Analysis division, resulting in a partial write failure. Specifically, one of the 88 RAID groups Lustre OST (OST0029) was not writable.
- The recovery work started at around 14:00 on Tue 18 Jun and finished at around 20:00.
- However, at 20:00, it was confirmed that some compute nodes were not able to access OST0029 (neither read nor write). Specifically, the following computation nodes.
- at017,at025,at026,at028,at029,at030,at031,at032,at033,at034,at035,at036,at037,at043,at044,at045,at046,at047,at048,at050,at051,at052,at053,at054,at055,at057,at058,at059,at060,at061,at062,at063,at064,at073,at074,at083,at084,at085,at087,at090,at095,at096,at097,at098,at099,at100,at101,at102,at103,at126,at127,at128,at129,at130,at131,at132,at133,at134,at135,at136 (60 of 136 Thin compute nodes Type 1a, AMD EPYC 7501 CPU)
- at139,at140,at141,at142,at143,at144,at145,at146,at147,at148,at149,at150,at151,at152,at153,at154,at155,at156,at157,at159,at160,at161,at162,at163,at164 (25 of 28 Thin compute nodes Type 1b, AMD ROMA CPU)
- it001,it002,it004,it006,it007,it008,it009,it010,it013,it014,it015,it017,it024,it025,it026,it027,it028,it029,it031,it032,it034,it035,t036,it040,it041,it048,it049,it050,it051,it052 (30 of 52 Thin compute nodes Type 2a, Intel CPU)
- igt001,igt003,igt005,igt006,igt007,igt008,igt011,igt012,igt013,igt014 (10 of 16 Thin compute nodes Type 2b, Intel CPU)
gw.ddbj.nig.ac.jp,gw2.ddbj.nig.ac.jp(all of 2 gateways for the general analysis division)- m01,m02,m03,m04 (4 of 10 medium nodes)
- dtn4 (one of 4 data transfer nodes used for DDBJ services)
-
- Wed 19 Jun, these nodes will be restarted in sequence to normalise access to Lustre7.
- Since two gateways in the general analysis section are included, login to the supercomputer will not be possible during this work and it will be disconnected from SSH communication.
- There are two gateways in the general analysis division, so if you cannot log in to one of them, please log in from the other one.
- Wed 19 Jun, these nodes will be restarted in sequence to normalise access to Lustre7.
Scope of impact
- From around 18:20 on 17 June to 14:00 on 18 June, any writing to OST0029 from all compute nodes is not possible. From around 14:00 to 20:00 on 18 June, any writing and reading to OST0029 from all compute nodes is not possible. The compute nodes listed above were also not read/write accessible as of 19 June. Please check your calculation results for any anomalies. (Jobs that do not use OST0029 are not affected, but whether or not using OST0029 is randomly determined.)
- The personal genome analysis division will not be affected.
- Communication breakdowns will occur for DDBJ services that use data transfer nodes dtn2 and dtn4.