BadgerDB database is too big for dataset

[solracsf]

There is a problem somewhere (maybe GC not running as it should?) with BadgerDB.
In a production env. the database is almost 10Gb in size:

# du -sh /opt/badger
9.2G    /opt/badger

But after doing a :

# juicefs dump badger:////opt/badger /tmp/latest.json
Scan keys: 7041370/7041370 [===========================================================]  136371.9/s used: 51.633579922s
Dumped entries: 2551157/2551157 [===========================================================]  48629.4/s  used: 52.461271285s

# juicefs load badger:///tmp/badgertest /tmp/latest.json
Loaded entries: 2551157/2551157 [===========================================================]  32701.7/s used: 1m18.01287635s

We can observe that the new database is 96% smaller !!

# du -sh /tmp/badgertest
303M    /tmp/badgertest

Also, doing a backup using badger CLI:

# badger backup -f /tmp/badger.dump --dir /opt/badger

# du -sh /tmp/badger.dump
494M    /tmp/badger.dump

• JuiceFS 1.1.0+2023-09-04.08c4ae6

[solracsf]

In another instance (running for almost a year) also using Badger, i can confirm this problem by dumping and loading the dump:

# du -sh /opt/badger
6.4G   /opt/badger

# juicefs dump --keep-secret-key badger:///opt/badger /tmp/meta-dump.json
Scan keys: 38127/38127 [==============================================================]  255614.4/s used: 149.17406ms
Dumped entries: 15472/15472 [==============================================================]  171927.0/s used: 90.011918ms

# juicefs load badger:///opt/badger_new /tmp/meta-dump.json                                                     2023/11/23 10:49:07.989729
Loaded entries: 15472/15472 [==============================================================]  104029.6/s used: 148.735698ms

# du -sh /opt/badger_new
1.5M    /opt/badger_new

[davies]

The garbage collection is triggered in every hour, so there should be other issues.

Can you list the sizes of all the files in badger directory?

[solracsf]

1.1G    000075.vlog
1.1G    000076.vlog
1.1G    000077.vlog
1.1G    000078.vlog
1.1G    000079.vlog
1.1G    000080.vlog
276M    000081.vlog
1.1G    000082.vlog
1.1G    000083.vlog
548M    000084.vlog
4.6M    000656.sst
68K     000668.sst
7.8M    000779.sst
4.5M    000782.sst
7.8M    000783.sst
7.8M    000787.sst
7.8M    000811.sst
896K    000816.sst
7.8M    000818.sst
2.3M    000819.sst
7.8M    000820.sst
3.5M    000822.sst
7.8M    000843.sst
4.0M    000846.sst
1.7M    000857.sst
28K     000876.sst
392K    000878.sst
7.8M    000880.sst
7.8M    000882.sst
7.8M    000883.sst
7.8M    000885.sst
7.8M    000886.sst
4.0K    000887.sst
1.3M    000888.sst
7.8M    000889.sst
7.8M    000891.sst
3.5M    000892.sst
7.8M    000893.sst
7.8M    000896.sst
200K    000897.sst
7.8M    000913.sst
7.8M    000914.sst
7.8M    000915.sst
7.8M    000916.sst
7.8M    000917.sst
7.8M    000918.sst
7.8M    000919.sst
1.3M    000920.sst
7.8M    000921.sst
7.8M    000922.sst
7.8M    000923.sst
7.8M    000924.sst
3.5M    000925.sst
7.4M    000926.sst
7.8M    000927.sst
7.1M    000928.sst
7.8M    000929.sst
204K    000930.sst
1.8M    000931.sst
444K    001970.sst
900K    001971.sst
3.9M    001972.sst
4.0M    001973.sst
3.9M    001974.sst
3.7M    001975.sst
4.1M    001976.sst
64K     001977.sst
2.1M    001978.sst
3.8M    001979.sst
20M     001980.sst
49M     00471.mem
4.0K    DISCARD
4.0K    KEYREGISTRY
4.0K    LOCK
36K     MANIFEST

[davies]

It looks like badger can't compact the vlog file as expected, can you file an issue to badger?

[solracsf]

Seems similar to hypermodeinc/badger#1995 and hypermodeinc/badger#2003 so let's keep discussion upstream.