Fix https://github.com/seaweedfs/seaweedfs/issues/3714
The destination chunks may be empty. For example, the file is updated and the volume is vacuumed. In this case, the sync would miss the old chunks. This is fine. However, the entry would have correct metadata but missing chunks.
For this case, the simple metadata comparison would be wrongly skipping data changes, and the file will stay empty unless file content md5 is changed.
* remove old raft servers if they don't answer to pings for too long
add ping durations as options
rename ping fields
fix some todos
get masters through masterclient
raft remove server from leader
use raft servers to ping them
CheckMastersAlive for hashicorp raft only
* prepare blocking ping
* pass waitForReady as param
* pass waitForReady through all functions
* waitForReady works
* refactor
* remove unneeded params
* rollback unneeded changes
* fix
Run two servers with volumes and fillers:
server -dir=Server1alpha -master.port=11000 -filer -filer.port=11001 -volume.port=11002
server -dir=Server1sigma -master.port=11006 -filer -filer.port=11007 -volume.port=11008
Run Active-Passive filler.sync:
filer.sync -a localhost:11007 -b localhost:11001 -isActivePassive
Upload file to 11007 port:
curl -F file=@/Desktop/9.xml "http://localhost:11007/testFacebook/"
If we request a file on two servers now, everything will be correct, even if we add data to the file and upload it again:
curl "http://localhost:11007/testFacebook/9.xml"
EQUALS
curl "http://localhost:11001/testFacebook/9.xml"
However, if we change the already existing data in the file (for example, we change the first line in the file, reducing its length), then this file on the second server will not be valid and will not be equivalent to the first file
Снимок экрана 2022-02-07 в 14 21 11
This problem occurs on line 202 in the filer_sink.go file. In particular, this is due to incorrect mapping of chunk names in the DoMinusChunks function. The names of deletedChunks do not match the chunks of existingEntry.Chunks, since the first chunks come from another server and have a different addressing (name) compared to the addressing on the server where the file is being overwritten.
Deleted chunks are not actually deleted on the server to which the file is replicated.
Running mount outside of the cluster would not need to expose all the volume servers to outside of the cluster. The chunk read and write will go through the filer.
chrislu
Ryan Russell
askeipx
Konstantin Lebedev