new random introducer port on each start

[T_X]

The patch from #1385, commit 2d16a16ee3d99482, does not seem fully fix the issue from e5c4e83f4cfe3769 (or introduces a new issue): When starting the introducer, it always sets a new random port in both introducer.furl and introducer.port (without any complaints though).

I've been using 'darcs obliterate' and 'darcs pull' to track this down to commits e5c4e83f4cfe3769 + 2d16a16ee3d99482, have been 'darcs obliterat'ing until 420e3aa2930cf1e7. This commit still works. Applying e5c4e83f4cfe3769 + 2d16a16ee3d99482 on top of 420e3aa2930cf1e7 does not seem to work anymore and introduces this random port issue for me.

[davidsarah]

Replying to T_X:

The patch from #1385, commit 2d16a16ee3d99482, does not seem fully fix the issue from e5c4e83f4cfe3769 (or introduces a new issue): When starting the introducer, it always sets a new random port in both introducer.furl and introducer.port (without any complaints though).

Please attach the tahoe.cfg for the introducer, and a listing of the introducer base directory.

[T_X]

tahoe.cfg of the buggy introducer node

[T_X]

Listing of the introducer node's base directory:

$ ls -lah /home/mesh-node/.tahoe-intro
total 40K
drwxr-xr-x  5 mesh-node mesh-node 4.0K Aug  4 20:39 .
drwxr-xr-x 23 mesh-node mesh-node 4.0K Aug  4 20:39 ..
-rw-------  1 mesh-node mesh-node   84 Aug  4 20:39 introducer.furl
-rw-------  1 mesh-node mesh-node    6 Aug  4 20:39 introducer.port
drwxr-xr-x  3 mesh-node mesh-node 4.0K Jul 29 15:27 logs
-rw-r--r--  1 mesh-node mesh-node   33 Aug  4 20:39 my_nodeid
drwx------  2 mesh-node mesh-node 4.0K Jul 29 15:27 private
-rw-r--r--  1 mesh-node mesh-node  520 Jul 29 15:27 tahoe.cfg
-rw-r--r--  1 mesh-node mesh-node  311 Jul 29 15:27 tahoe-introducer.tac
drwxr-xr-x  2 mesh-node mesh-node 4.0K Jul 29 15:27 tmp

[davidsarah]

node.py: ensure that client and introducer nodes record their port number and use that port on the next restart, fixing a regression caused by #1385. fixes #1469.

[davidsarah]

test_runner.py: test that client and introducer nodes record their port number and use that port on the next restart. This tests for a regression caused by ref #1385.

[davidsarah]

Includes fix, tests, and cleanups of the test code.

[davidsarah]

The problem was that e5c4e83f4cfe3769, in addition to removing the code that read the old config files, removed the code that read config.port (for a client node) or introducer.port (for an introducer).

[zooko]

Good patch! The only thing I might possibly do different is that this method named ​_remove is redundant with (but probably better than) fileutil.remove_if_possible​. Anyway: +1

[david-sarah@…]

In b9121e045fe2894c:

test_runner.py: fix a race condition in the test when NODE_URL_FILE is written before PORTNUM_FILE. refs #1469

[david-sarah@…]

In 05684b9251780046:

node.py: ensure that client and introducer nodes record their port number and use that port on the next restart, fixing a regression caused by #1385. fixes #1469.

[davidsarah]

Replying to zooko:

Good patch! The only thing I might possibly do different is that this method named ​_remove is redundant with (but probably better than) fileutil.remove_if_possible​. Anyway: +1

I changed it to use fileutil.remove, which is exactly what is needed. (Fail if the file doesn't get removed, but succeed if it didn't exist to start with or was removed concurrently.)