unixwitch/jobserver

Flexible replacement for 'cron' and 'at'

jobserver's Issues

ACLs

Minimal framework for ACLs is available since 4aead7d, but ACLs aren't actually supported. Need some commands like this:

$ job access 13 user:river:vmsd
$ job access global group:users:v---

Mails could be a little clearer

When a job exit action is set to 'disable', the mail just says 'The job has been disabled.' It should probably explain why.

Add a 'max-time' rctl

A new rctl that would limit wall clock time of a job. Stop method should be invoked automatically once the limit is reached.

$ job limit 13 max-time 600

Command to list upcoming jobs

Need a command to list scheduled jobs in execution order, with time of execution.

Configurable log rotation behaviour

Currently log rotation is hard-coded to 1MB filesize and 2 keeps. It should be possible to configure this, or to disable log rotation entirely.

Remove $Id$ tags

Allow umask to be set for a job

"Maintenance" should be an rstate, not a state

Merge 'set' and 'limit'

There's no point having different commands for this. Make 'job set x max-data-size' work and remove 'job limit' and 'job unlimit'.

Make log file format configurable

The logfile format should be configurable, preferable separately for continuous and scheduled jobs.

Allowed scheduled jobs to be enabled or disabled

Currently scheduled jobs cannot be enabled or disabled; a scheduled job is always run. It should be possible to disable a scheduled job, retaining the schedule but not running it.

Need 'job set name'

Like 'set fmri', but it sets the job:/username/ automatically.

Add read timeouts to networking

Add 'maillog' action

This would be like 'mail', but it would include the last few lines of the logfile.

Scheduling a job should enable it

Don't kill all jobs when restarting the jobserver

Instead, associate our job FMRI with the process contract, and re-attach when the server restarts.

Clean up contract handling

... and move it into a separate contract_t wrapper.

"job quota" is broken

% ./job quota jobs-per-user 25
Invalid parameter.

'job show' is missing a newline if there are no limits

Add a way for 'job limit' to print raw units

Sometimes raw units may be wanted, e.g. for scripting. Add a switch to 'job limit' to disable formatting.

'job show' should list time until next scheduled execution

Document exit actions in job_set(1)

Store job history

There should be a way to view job history, e.g. when it was started, if it failed, changes to it, etc.

"job limit" is inconsistent about units

% ./job limit 13 max-data-size      
max-data-size = 10240
% ./job limit 13               
max-data-size = 10.00MB

Start PAM session when running a tool

To enable account expiry to work, we need to start a PAM session when starting a tool.

Make 'mail' exit action work

Currently this does nothing. It should be implemented.

Don't let 'job set' set an invalid project

Currently the server doesn't validate project names in CHNG until the job actually executes. To prevent confusing users, it should validate it at CHNG time and return an error.

'disable' doesn't make sense for scheduled jobs

The default action for scheduled jobs is 'disable', but disable is actually ignored for scheduled jobs, so this doesn't do anything. Add a new 'reschedule' action, and make 'disable' work as expected.

'every minute' is a little odd

'every minute' sched just returns now+60, instead of calculating the top of the next minute like it should.

Need a way to do a clean shutdown

Currently when the jobserver exits, it doesn't end running jobs, which means they get SIGKILL instead of normal shutdown behaviour. The jobserver should handle SIGTERM and do a clean shutdown.

Need a way to pass job start errors back to the job server

Currently, a failure between forking and opening the logfile is not reported to the user. There should be a mechanism to allow this, perhaps by opening an FD_CLOEXEC pipe that errors would be reported over.

Ignore errors during stop

If a fail/crash occurs during job stop, ignore it.

Use FMRIs instead of ids/names

Instead of using an id and a name, jobs should be identified by a unique FMRI of the form job://

Mail should send a Date header

'job show' should print actual user not, name uid

Should try more than once to start a job

If a job fails to stay up (runs less than SCHED_MIN_RUNTIME) we should try to start it several times before placing it in maintenance, in case it was a transient problem.

Need 'job set fail/crash/exit'

Currently behaviour of fail/crash/exit is not configurable. It should be possible for the user to select 'restart' and/or 'mail' actions for these, e.g.:

job set 4 crash=restart,mail

Need a command to unset properties

Currently, unsetting a property is done by setting it to a null value. There should be a 'job unset' command instead.

Support scheduling at absolute time

Need to allow 'at'-style scheduling, e.g. 'job sched 1 at 13:00' or 'job sched 1 at 2009-01-01 23:00', or even 'job sched 1 in 5 minutes'.

"job disable" on a scheduled job prints misleading error

% ./job list
  ID NAME      USER     STATE       RSTATE    CMD
  13 test.sh   rriver   scheduled   stopped   /home/rriver/test.sh
% ./job disable 13
Cannot enable a scheduled job.

Add an nvlist interface to kvdb

So that nvlists can be automatically packed/unpacked, rather than doing it manually.

Need a way to stop a scheduled job

Currently it is impossible to stop a scheduled job. 'job disable' is disallowed, and 'job unschedule' unscheds the job but doesn't stop it. We probably need a 'job stop' command.

"job sched" help has outdated list of sched specs

'job show' should list job limits

Add 'job unsched -s' to stop the job as well as unscheding it

Need better error reporting

Currently we only use errno to report errors, which is often too vague. Add some negative errno values to describe jobserver-specific failures.