Add some more ctdb related hacks #124
Conversation
Add the cluster_meta_to_nodes function that will simply take a clustermeta object as an input and write out a corresponding nodes file for ctdb. Signed-off-by: John Mulligan <[email protected]>
Add an archive_tdb function that archives (moves) the tdb files that converted into ctdb backing files. Signed-off-by: John Mulligan <[email protected]>
Add a --write-node option to the ctdb-must-have-node command so that containers without a shared file system can at least join a ctdb cluster. This a bit hacky but it works to test things out. Signed-off-by: John Mulligan <[email protected]>
Passing this argument with a path will archive any of the tdb files that get migrated to ctdb to the specified path. This moves files from the typical location to aid in debugging (it's more obvious that this instance is using ctdb) and for comparison in case something needs to be compared post-conversion. Signed-off-by: John Mulligan <[email protected]>
|
cc @avanthakkar |
| nffh.flush() | ||
| os.fsync(nffh) |
There was a problem hiding this comment.
Could you please remind me why we need this pattern of doing both flush and os.fsync ?
It is repeated in this module; maybe have a local helper-function with minimal doc-string.
There was a problem hiding this comment.
To be fair it is a copy-paste that was tweaked. The goal of the pattern is to call flush to make sure any buffers internal to the process get flushed to the OS. Then fsync tells the os to flush any buffers it has and ensure the entire file is written to the storage device. Normally, flush is not required as it is done also by close, but to use fync we need an fd - so we end up having to manually flush first.
There was a problem hiding this comment.
Fair enough. I still think it would be nice to have it local helper-function but this can wait to future cleanups PR.
There was a problem hiding this comment.
Yeah, if you note I'm not writing very many tests for some of this stuff either. Normally, I'd criticize myself for that but I'm doing a bunch of prototyping simultaneously on the ceph codebase and sambacc. I'm trying to get as many of these into the codebase here in sambacc so I don't end up with massive private branches on both sides. :-)
I'd welcome anyone to jump in and help contribute tests and cleanups on top. :-)
| for tdbfile in _SRC_TDB_FILES: | ||
| for parent in _SRC_TDB_DIRS: | ||
| tdb_path = os.path.join(parent, tdbfile) |
There was a problem hiding this comment.
Instead of double nested for loop, consider something like:
from itertools import product
from pathlib import Path
for tdb_path in [ Path(p) / f for p, f in product(_SRC_TDB_DIRS, _SRC_TDB_FILES) ]:
...
There was a problem hiding this comment.
This is how the older, migrate functions is written. I don't want to deviate from the pattern now.
| nffh.flush() | ||
| os.fsync(nffh) |
There was a problem hiding this comment.
Fair enough. I still think it would be nice to have it local helper-function but this can wait to future cleanups PR.
Add some additional options to the ctdb related sambacc commmands. The --write-node option in particular is a bit hacky but it helps me get on with prototyping samba+ctdb in containers under ceph.