Replicate objects in case of HDD failure on another node #179

New issue

Open

opened 2023-03-27 15:15:36 +00:00 by fyrchik · 1 comment

fyrchik commented

2023-03-27 15:15:36 +00:00

Owner

From https://github.com/nspcc-dev/neofs-node/issues/2085

Head requests done by policer read information only from the metabase.
However, if data disk failed, the object could become unavailable.

From https://github.com/nspcc-dev/neofs-node/issues/2085 Head requests done by policer read information only from the metabase. However, if data disk failed, the object _could_ become unavailable.

fyrchik added the

discussion

triage

labels 2023-03-27 15:15:36 +00:00

fyrchik commented

2023-03-27 15:20:04 +00:00

Author

Owner

Possible solutions:

Provide a flag to HEAD operation, requiring node to consult main storage. Lot's of negatives here and we do not really solve the problem: the only way to know if the object is available is to actually read it.
Answer with a special status code if the shard is in read-only mode. This way we only read shards mode, but still it can lead to a lot of false positives: not every read-only shard has bad disks.
Do nothing. Seems ok to me: read-only shard is an exceptional situation, and evacuating data should be done by the administrator.

Possible solutions: 1. Provide a flag to HEAD operation, requiring node to consult main storage. Lot's of negatives here and we do not really solve the problem: the only way to know if the object is available is to actually read it. 2. Answer with a special status code if the shard is in read-only mode. This way we only read shards mode, but still it can lead to a lot of false positives: not every read-only shard has bad disks. 3. Do nothing. Seems ok to me: read-only shard is an exceptional situation, and evacuating data should be done by the administrator.

fyrchik added this to the vNext milestone 2023-05-18 08:33:16 +00:00