onlyjob
This is just an idea. I have a storage class with most chunks (60 million, over 6 times the count of chunks in all other storage classes combined). That storage...
Observerd the problem on 3.0.111 and previous releases: __I/O on MooseFS mount is blocked while chunkserver is scanning__. We have two (heavy) archival chunkservers with ~30 million chunks each. They...
Situation: a chunkserver with 8 HDDs, 7 of them are 1% full (newly installed) and one HDD is 99% full. (Disks were replaced after spectacular failure of disk enclosure that...
In two separate incidents I've observed incorrect handling of out of disk space condition in master. At first clients stopped responding and quick investigation revealed runaway log on Master server...
Chunk size is not optimal for small files (e.g. sources). Because chunk files are not sparse HDD space utilisation is not optimal and a lot of space is wasted. To...
Situation: chunkserver with 30 million chunks and several hard disk is re-started with `HDD_HIGH_SPEED_REBALANCE_LIMIT = 3`: This is what happens after scanning of local HDDs: ``` Jan 31 09:01:19 mfschunkserver[103002]:...
When a _storage class_ definition is changed, replication begins but there is no indication of how many chunks are out of place in the "CGI/Resources" page. It would be very...
In `mfshdd.cfg` a disk is marked as `
When HDD runs out of inodes, chunkserver is getting stuck with `` load. Perhaps such situation could be handled gracefully, with reported 100% HDD usage. Reproducing should be easy on...
Promoting metalogger's data to master is needlessly difficult because master should be started with `-a` option exactly once to convert `_ml` files. Metalogger data conversion should have been implemented as...