Multi instance container setup and repo access

[codeplay87]

I am trying to understand how opengrok mirroring and indexing works to validate that it won't cause any issues if we have multiple docker instances pointing at shared volume.

We currently have the following setup in k8s on azure:

• Persistent Volume for the opengrok (src, data, etc)
• Persistent Volume for the tomcat server config.xml

If we had n* of instances running against the same shared data does opengrok do any locking of those repos when it runs its mirroring or indexing?

Many thanks in advance.

[vladak]

The OpenGrok tooling (opengrok-mirror and opengrok-sync) use file based locking in system temporary directory (usually /tmp in Linux distros). If projects are enabled, the lock files will be specific to given project.

As for the indexer itself, one can set the Lucene locking style with the --lock option. The default is off, i.e. no locking. The other choices are either simple or native which corresponds to Lucene's SimpleFSLockFactory and NativeFSLockFactory, respectively. I believe the former will use lock files, while the latter will employ fcntl() style locking (on Unix-like systems).

[codeplay87]

Thanks for getting back so quickly. Is there anyway to set the location of that directory used for the file based locking for mirror and sync?

[vladak]

Not at the moment, e.g. for opengrok-mirror it explicitly uses tempfile.gettempdir():

opengrok/tools/src/main/python/opengrok_tools/mirror.py

Lines 212 to 214 in f075707

	lock = FileLock(os.path.join(tempfile.gettempdir(), lockfile + ".lock"))
	try:
	with lock.acquire(timeout=0):

Under the covers, filelock will use whatever implementation it can find, preferring native (alluding to the Lucene locking), i.e. fcntl() based locking on Unix-like systems or if it is not found, it will fall back to lock files (called "soft locking" there): https://github.com/tox-dev/py-filelock/blob/dbe3ba1470838f4aded968c8d364d6e8ab6c2c00/src/filelock/__init__.py#L27-L31