2024-11-15T02:50:36.578Z

<Patrick Donnelly> define "size of the FS"

2024-11-15T09:43:26.231Z

<Venky Shankar> [1] During search if a reg file is seen, it synchronized at that point, but for directory, its recursively listed and the entries are added to the sync stack.

2024-11-15T09:43:50.295Z

<Venky Shankar> [2] yeh -- @jcollin found it a bit more easy to use a queue when syncing using snap diff.

2024-11-15T09:44:36.646Z

<Venky Shankar> [3] and [4]: it depends, but yes, you'll have to keep a bit more entries in the memory when encountering dirs and using a stack...

2024-11-15T10:10:25.160Z

<Md Mahamudur Rahaman Sajib> @Venky Shankar
Thanks, for your insights. Then my question is why can't we add parallelism for single directory.
If are doing bfs, let's say just assume we are taking snapshot for the first time(The solution should also work for inceremental approach as well), we can a maintain a task queue in the thread pool.

In a naive level a task will represent,
task: (create a directory, push tasks of creating subdirectories/files in the task queue)

Why can't we use this bfs-ish approach parallelly. In that case subdirectries which don't have ancestor-descendent relationship always go in parallel. I know there is a lot of details need to care about, but I am thinking  on a high level.

Actually we are having performance issue with customer. So just want to learn a bit.

2024-11-15T12:14:04.537Z

<Markuze> Maximum size in bytes.
```Filesystem      Size  Used Avail Use% Mounted on
/dev/dm-0       476G  397G   75G  85% /```

2024-11-15T20:58:32.922Z

<Ivveh> hmm is there no ceph-common 18.2.4 for ubuntu 24.04?

2024-11-15T21:09:58.386Z

<Patrick Donnelly> by default cephfs will use all available space in RADOS

2024-11-15T21:10:04.557Z

<Patrick Donnelly> you would need to limit it with cephfs quotas

2024-11-15T21:19:57.914Z

<Markuze> Yeah, I figured that.
I was wondering about the 99G figure i see on ceph mounts on a vstart claster