connections aren't closed on CAT_INVALID_USER

[bh9]

metalnx doesn't always close failed connections immediately. Our dev irods database was effectively DOS'd by metalnx creating too many connections. A lot (possibly all) of the connections which don't get cleared up have fail with CAT_INVALID_USER, something like

Dec  9 05:39:18 pid:1948 NOTICE: Agent process 21497 started for puser=rods and cuser=rods from 172.27.84.193
Dec  9 05:39:18 pid:21497 NOTICE: rsAuthCheck: chlCheckAuth status = -827000
Dec  9 05:39:18 pid:21497 ERROR: [-]	iRODS/server/api/src/rsAuthResponse.cpp:74:rsAuthResponse :  status [CAT_INVALID_USER]  errno [] -- message []
Dec  9 09:22:56 pid:1948 NOTICE: Agent process 21497 exited with status 24576

note that a connection which has clearly failed is open for 4 hours, when it should have been closed immediately.

[bh9]

Worth noting the CAT_INVALID_USER was triggered by the fact that I didn't notice the jobs config, so as a side point, could the entries in metalnx.properties be documented please?

[trel]

jobs are now gone since #214... so perhaps this is moot?

[JustinKyleJames]

jobs are now gone since #214... so perhaps this is moot?

I believe this is still an issue with the login page. When logging in using valid credentials, the ips does not show a hung agent connected from jargon. I see the same when navigating around.

When trying to login with an invalid user, I see the agent alive with jargon.

Every 5.0s: ips                                                                                                                                                28a3e492c919: Mon Jan  6 20:54:59 2025

Server: irods-catalog-provider
    28478 dne#tempZone  0:00:53  jargon  172.18.0.4
    28711 rods#tempZone  0:00:00  ips  172.18.0.2

[trel]

ah, good find.

[JustinKyleJames]

I finally found what was causing the CAT_INVALID_USER to leave connections open. When I updated the following code to catch InvalidUserException as well as AuthenticationException the connection closed as expected.

Note that prior to making this change the agent stayed open for a while but garbage collection would clean up that connection when it ran. Now it will clean it up immediately.

https://github.com/DICE-UNC/jargon/blob/5076fce55d858a6e2431fcbdf3a78b34a6af5f20/jargon-core/src/main/java/org/irods/jargon/core/connection/AbstractIRODSMidLevelProtocolFactory.java#L98-L102

[trel]

Full explanation of behavior and fix. Excellent.

[trel]

complete - closing.