Implement MountSnapshotJob / UnmountSnapshotJob

[mkpankov]

Signed-off-by: Michael Pankov work@michaelpankov.com

---

This change is 

[mkpankov]

It works like this:

[root@mds1 vagrant]# lctl snapshot_list -F zfsmo 

filesystem_name: zfsmo
snapshot_name: test1
snapshot_fsname: 1494052f 
create_time: Mon Jul 27 14:09:53 2020
modify_time: Mon Jul 27 14:09:53 2020
status: not mount

➜  ~ jarg message="Test command" 'jobs[0][class_name]=MountSnapshotJob' 'jobs[0][args][hosts][0]=/api/host/1/' 'jobs[0][args][fsname]=zfsmo' 'jobs[0][args][name]=test1' | http --verbose --json --verify=false https://localhost:8443/api/command/ authorization:'ApiKey api:15ec8aca8b4139ad741e4bbf9157b1e193a07a52'
POST /api/command/ HTTP/1.1
Accept: application/json, */*
Accept-Encoding: gzip, deflate
Connection: keep-alive
Content-Length: 146
Content-Type: application/json
Host: localhost:8443
User-Agent: HTTPie/1.0.3
authorization: ApiKey api:15ec8aca8b4139ad741e4bbf9157b1e193a07a52

{
    "jobs": [
        {
            "args": {
                "fsname": "zfsmo",
                "hosts": [
                    "/api/host/1/"
                ],
                "name": "test1"
            },
            "class_name": "MountSnapshotJob"
        }
    ],
    "message": "Test command"
}

HTTP/1.1 201 Created
Connection: keep-alive
Content-Length: 212
Content-Type: application/json
Date: Thu, 13 Aug 2020 10:59:33 GMT
Location: /api/command/30/
Server: nginx/1.16.1
Vary: Accept
X-Frame-Options: SAMEORIGIN

{
    "cancelled": false,
    "complete": false,
    "created_at": "2020-08-13T10:59:33.597354",
    "errored": false,
    "id": 30,
    "jobs": [
        "/api/job/42/"
    ],
    "logs": "",
    "message": "Test command",
    "resource_uri": "/api/command/30/"
}

[root@mds1 vagrant]# lctl snapshot_list -F zfsmo 

filesystem_name: zfsmo
snapshot_name: test1
snapshot_fsname: 1494052f 
create_time: Mon Jul 27 14:09:53 2020
modify_time: Mon Jul 27 14:09:53 2020
status: mounted

[mkpankov]

@jgrund dependencies and locks are missing, but I wanted to clarify if the approach taken is correct with respect to "DB calls should be made in Rust" as specified in the issue.

Since the state machine stays in Python and we need to save the Job itself, I guess these don't count as DB calls?

[jgrund]

Since the state machine stays in Python and we need to save the Job itself, I guess these don't count as DB calls?

Yes, I'm more talking about associated queries needed to get related objects.

[mkpankov]

I've tested this with non-imported pools and no dependencies were necessary besides ldev.conf.

[jgrund]

I've tested this with non-imported pools and no dependencies were necessary besides ldev.conf.

At the very least the filesystem needs to be in the available state. Mounting a snapshot should not work if the fs is unavailable I think

[mkpankov]

So, I've retested everything.

1. Mounting existing snapshot doesn't actually require running lustre FS. It works after ZFS pools are imported with import-pools.
2. With DependOn(ManagedFilesystem.objects.get(name=self.fsname), "available") as in current version, it does require a running lustre FS that is detected by IML.

[jgrund]

We'll want to check if this is needed for ldiskfs since it's apparently not needed with zfs

[johnsonw]

I think this is fine for now but in the future, we will have a local action plugin that we can call to get the fqdn of the MGS node.

[johnsonw]

Not really a big deal but the long descriptions are shorter than the standard descriptions. You could store the values that you have under description in this help.py file and then call them:

 @classmethod
    def long_description(cls, stateful_object):
        return help_text["unmount_snapshot"] % stateful_object.fqdn

    def description(self):
        return self.long_description(self)