The London Perl and Raku Workshop takes place on 26th Oct 2024. If your company depends on Perl, please consider sponsoring and/or attending.
 / 
Dicop-Server-3.03
++
/ BUGS

BUGS - Only a dead Bug is a good Bug

External modules

Net::Server

When using v0.85 of Net::Server, starting dicopd may emit a warning from Net::Server along the lines of:

        Couldn't POSIX::setuid to \"$uid\" []

This seems to be spurious and can be ignored.

Server

  • When a job is still only a few hours old, all of it's chunks will be sized 5 minutes. However, other jobs have a chunk size of 35, 55 or any other value, depending on what the clients are set to. The algorithmn to hand out chunks, however, does not take these sizes into account, it merely distributes chunk counts. This means a job with 5 minute chunks will get 90% of all chunks, but not 90% of all CPU time.

  • Maybe the config should not be viewable by everyone who can see the cmd_status, since it details the allowed/disallowed status/stats/work/admin settings. OTOH, even when you can "see" the data, you still don't have admin rights because you don't know the right password...

  • The server should deny any DONE/SUCCESS reports of the client with a failure counter too high. (on the line of reasoning that they are probably wrong/bogus). This will only happen if a client has failures _and_ gets more than one chunk at a time from the server (otherwise the client can't report chunks to the server because it isn't getting them in the first place - it is denied work while having a high failure counter) - or if a client sends in bogus reports all the time.

  • Server leaves out testcases for workers it does not have. (f.i. a worker for architecture foo is missing, and a client working in foo requests tests) However, the actual work from a job may still be sent to the client, even though the client can not work on it. (Really? Check it!)

  • Config should accept comments at end of line, not only entire lines as comments

  • A client that sends a FAILED and a SUCCESS testcase for one jobtype might not get disabled (or accidentily re-enabled) when the FAILED testcase is handled first. (Server should handle failed testcases last)

  • Store what tests you send to which client. This fixes the bug that the server needs to rely on client sending back correct 'type' for report. (This falls under the "cache with the secret tokens as ID, and job/chunk as data" idea that would allow us to NOT send the client the job/chunk info at all)

General

  • Adding character sets with multi-byte "characters" (think unicode, or "words" as "characters" like qw/hans wurst/) is not yet possible. One way to solve this is to have a text field for input and split the input at linefeeds into "characters".

  • The server should send only one email containing all offline clients, not one mail for each client. This could result in a large number of emails being generated when a lot of clients go offline simultanously (network/power outage etc). Also, sending the offline email should probably be supressed when the server starts up (on the reasoning that if the server was down, the clients will also be offline, but you know this already)

  • The target dir must be called "target", otherwise the client can't ask the server for a download location (hardcoded check on /^(worker|target)$/) and also the worker would not find the JOBID.set files (hardcoded check for file in ../../target/.