王晓哲 wrote:
Hi all,
I have tried MogileFS for a while, and did the following things:
1. add a domain and a class in that domain, with mindevcount=3 (make 3
replications);
2. insert a new file into MogileFS, and I saw there are 3 storage nodes
(node A,B,C) contained that file according to 'mogadm stats';
3. manually mark one of those nodes (node A) to be down, and try to
retrieve that file: file retrieved normally, and a new replication is
made up on another node (node D) as expected;
4. manually mark the previous down node alive, and delete the file.
Were you marking it as down, or dead?
If _down_, the correct behavior is to retry the delete later (since it
could not succeed in removing all of the files). Do you have any rows in
file_to_delete_later? Also, I don't _think_ the replicator would
re-replicate files if a host or device is simply marked as down. When
you mark one as dead the reaper process will do an immediate fixup though.
Here comes the problem: after the file has been deleted, there are no
contents in node B,C and D, but still left one copy of the file in node
A (can be seen through 'mogadm stats'). That copy could not be accessed
anymore, but do occupied the storage spaces. Even though I tried to
clean up the whole storage directory on node A, the mogadm file
statistics remain unchanged.
With a long running systems and lots of files, this phenomenon will be
very annoying. So is there any formal method to clean up all of these
dangling files in a MogileFS system?
Yeah I don't really see this. I don't try too hard to find it, but if
it's happening I'd consider it a bug.
-Dormando
