hi Bruce,At Chuck's suggestion, I've added an initial PyNFS test to aid work on a courteous server. A simple test, along the lines you indicated, that locks a file, waits twice the lease period, and tries to unlock:
OK -> PASS (courteous server) BADSESSION -> WARNING (discourteous server)Before sending my patch, Chuck asked me to add the second test you suggested:
- A second test creates a new client, acquires a file lock, and waits two lease periods. Then creates a second client, which attempts to acquire the lock. The second client should succeed. This doesn't seem to differentiate between these three cases:1. a discourteous server, which invalidates the client 1 state, and frees all client 1's locks, upon lease expiry, then allows client 2 to lock the file. The above test spec would result in a PASS for a discourteous server, which doesn't seem right.
2. a broad-grained courteous server, which invalidates the client 1 state, and frees all client 1's locks, because of conflicting access from client 2 (after client 1's lease expiry), who is then granted the lock. A PASS here would be correct.
3. a fine-grained courteous server, which persists the session, but revokes that particular client 1 lock, because of conflicting access from client 2 (after client 1's lease expiry), who is granted the lock. A PASS here would be correct.
Or am I misreading your suggestion?If I've read it right, the test could differentiate between cases 2) and 3), by having client 1 try to unlock, after client 2 successfully locks, where client 1 will see either BADSESSION (case 2) or SOME_STATE_REVOKED / EXPIRED (case 3). But we don't need to differentiate cases 2) and 3), since a PASS would be correct in either case.
However that won't differentiate between cases 1) and 2), where client 1 will see BADSESSION in each case. Yet case 1) ought to result in a WARNING, and case 2) in a PASS?
cheers, calum.
Attachment:
OpenPGP_signature
Description: OpenPGP digital signature
