Forums

Home / Forums

You need to log in to create posts and topics. Login · Register

bluestore_block_db_size ignored

Pages: 1 2

I am using PetaSAN 3.1. I was successful on three OSDs. The other three I will try when I have the next maintenance slot.

Here the value before

ceph daemon osd.6 perf dump | jq '{bluefs:.bluefs}'
{
"bluefs": {
"gift_bytes": 0,
"reclaim_bytes": 0,
"db_total_bytes": 794567872512,
"db_used_bytes": 42356150272,
"wal_total_bytes": 0,
"wal_used_bytes": 0,
"slow_total_bytes": 351843581952,
"slow_used_bytes": 66498461696,
"num_files": 1730,
"log_bytes": 9670656,
"log_compactions": 29,
"logged_bytes": 465784832,
"files_written_wal": 2,
"files_written_sst": 8760,
"bytes_written_wal": 187023945728,
"bytes_written_sst": 514929278976,
"bytes_written_slow": 0,
"max_bytes_wal": 0,
"max_bytes_db": 46491734016,
"max_bytes_slow": 0,
"read_random_count": 30085874,
"read_random_bytes": 707852574430,
"read_random_disk_count": 12452426,
"read_random_disk_bytes": 621318558952,
"read_random_buffer_count": 17751574,
"read_random_buffer_bytes": 86534015478,
"read_count": 6727080,
"read_bytes": 185227070725,
"read_prefetch_count": 6716055,
"read_prefetch_bytes": 184865965093,
"read_zeros_candidate": 0,
"read_zeros_errors": 0
}
}

and after compact

ceph daemon osd.6 perf dump | jq '{bluefs:.bluefs}'
{
"bluefs": {
"gift_bytes": 0,
"reclaim_bytes": 0,
"db_total_bytes": 794567872512,
"db_used_bytes": 92989788160,
"wal_total_bytes": 0,
"wal_used_bytes": 0,
"slow_total_bytes": 351843581952,
"slow_used_bytes": 0,
"num_files": 1362,
"log_bytes": 2453504,
"log_compactions": 0,
"logged_bytes": 1085440,
"files_written_wal": 1,
"files_written_sst": 0,
"bytes_written_wal": 44199936,
"bytes_written_sst": 0,
"bytes_written_slow": 0,
"max_bytes_wal": 0,
"max_bytes_db": 92989788160,
"max_bytes_slow": 0,
"read_random_count": 33987,
"read_random_bytes": 166101580,
"read_random_disk_count": 1540,
"read_random_disk_bytes": 47363313,
"read_random_buffer_count": 32448,
"read_random_buffer_bytes": 118738267,
"read_count": 2531,
"read_bytes": 143946259,
"read_prefetch_count": 2297,
"read_prefetch_bytes": 137845150,
"read_zeros_candidate": 0,
"read_zeros_errors": 0
}
}

 

I think the slow bytes used is the value that shows the spillover.

Very good, the manual compaction fixed it. Note that the OSD will regularly perform db compaction so it should resolved by itself.

Pages: 1 2