SyncTest socket reuse bug

[novakboskov]

SyncTest function when used with GenSync over CommSocket forks two pairs of GenSyncClient, GenSyncServer. Both pairs reconcile over the same socket. Since one of the pairs must finish first, the one behind fails to close the socket. It's already closed by that time.

I tried to depict the scenario (the socket is in blue):

The bug can easily be reproduced by running IBLTSyncTest or CuckooSyncTest with DEFAULT_LOGLEVEL=METHOD.

That prints the message Attempted closing of socket that is not connected to anything. from CommSocket::commClose().

The first pair of GenSyncClient, GenSyncServer closes the socket when done and sets CommSocket::my_fd to -1, the trailing pair of GenSyncClient, GenSyncServer fails to close the socket because CommSocket::my_fd == -1 by that time.

[arorashu]

Posting an update.

So, I found that this issue has been somewhat discussed in the BUGS.txt, and also in the code

[arorashu]

PR #61 is created for this issue. The architecture of syncTest is as follows:

[arorashu]

Actually, I have further simplified it now:

[arorashu]

I have written here semi-formally how syncTest is expected to behave:

For all syncs that implement GenSync, syncTest tests that the following guarantees hold:

Let Peer P1 be client with data - set A : GenSyncClient and Peer P2 be server with data - set B : GenSyncServer

Now, GenSyncClient synchroizes using GenSync :: clientSyncBegin, and GenSyncServer synchroizes using GenSync :: serverSyncBegin

At the end, GenSyncServer data is modified. The server is now expected to have the data - set A U B. For "one-way" syncs, the client should be unchanged, else the client is also modified to contain data - A U B

If the resultant data set for GenSyncClient and GenSyncServer is as expected, the test passes, else the test fails.

[trachten]

Looks right to me!

On Jun 19, 2020, at 8:34 AM, 6/19/20, Shubham Arora notifications@github.com wrote:

I have written here semi-formally how syncTest is expected to behave:

For all syncs that implement GenSync, syncTest tests that the following guarantees hold:

Let Peer P1 be client with data - set A : GenSyncClient and Peer P2 be server with data - set B : GenSyncServer

Now, GenSyncClient synchroizes using GenSync :: clientSyncBegin, and GenSyncServer synchroizes using GenSync :: serverSyncBegin

At the end, GenSyncServer data is modified. The server is now expected to have the data - set A U B. For "one-way" syncs, the client should be unchanged, else the client is also modified to contain data - A U B

If the resultant data set for GenSyncClient and GenSyncServer is as expected, the test passes, else the test fails.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/trachten/cpisync/issues/48#issuecomment-646611189, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACM5LU46ZI6XSFJJIILRILLRXNLMHANCNFSM4MFDJLHQ.

---

Ari Trachtenberg, Boston University http://people.bu.edu/trachten mailto:trachten@bu.edu

Random quote of the day: 
A day without sunshine is like, night. 

[arorashu]

This issue was fixed in PR#61. Closing here.