[bitbake-devel] [PATCH 1/2] V4 Disk space monitoring

[Robert Yang]

Hi Richard,

Thank you very much for your detailed review, I've fixed most of
them as you suggested except the following ones, and please see my
comments below.

On 02/23/2012 01:28 AM, Richard Purdie wrote:
> Hi Robert,
>
>>     # Set disk space and inode interval, the unit can be G, M, or K, but do
>>     # NOT use the GB, MB or KB (B is not needed), the format is:
>>     # "disk space interval, disk inode interval",  the default value is
>>     # "10M, 50" which means that it would warn when the free space is
>>     # lower than the minimum space(or inode), and would repeat the action
>>     # when the disk space reduces 10M (or the amount of inode reduces 50)
>>     # again.
>>     #BB_DISKMON_INTERVAL = "10M,10K"
>
> I'm wondering how useful this interval is? Surely once we've warned,
> aborted or stopped starting new tasks, running the action again isn't
> much use? This is particularly true with the change I'm proposing above.
>

I think there are 3 actions we can do: WARN, STOPTASKS or ABORT, this is
only useful for "WARN", when the action is "WARN", there would be too many
WARNINGS without the interval value.

>> +def errRet(info):
>> +    logger.error("%s" % info)
>> +    logger.error("Disk space monitor will NOT be enabled")
>> +    return None
>> +
>> +def errRetTwo(info):
>> +    logger.error("%s" % info)
>> +    logger.error("Disk space monitor will NOT be enabled")
>> +    return None, None
>
> These should be one multiline logger.error() call (using \n for
> newlines).
>

Yes, I've fixed this, also combine errRet and errRetTwo into one
function printERR.

>> +        for dev in devDict:
>> +            st = os.statvfs(devDict[dev][0])
>> +            # The free space, float point number
>> +            freeSpace = st.f_bavail * st.f_frsize
>> +            if devDict[dev][1] is not None and freeSpace<  devDict[dev][1]:
>> +                # Always show warning, and this is the default "WARN" action
>> +                if self.preFreeSpace[dev] == 0 or self.preFreeSpace[dev] - freeSpace>  self.dm.spaceInterval:
>> +                    logger.warn("The free space of %s is running low (%.3fGB left)" % (dev, freeSpace / 1024 / 1024 / 1024.0))
>> +                    self.preFreeSpace[dev] = freeSpace
>> +                if self.dm.action == "NO_NEW_TASK":
>> +                    logger.warn("No new tasks can be excuted since BB_DISKMON_ACTION = \"NO_NEW_TASK\"!")
>> +                    return 1
>> +                elif self.dm.action == "ABORT":
>> +                    logger.error("Immediately abort since BB_DISKMON_ACTION = \"ABORT\"!")
>> +                    sys.exit(1)
>
> Please don't do this. There is a way to immediately abort the runqueue
> by calling rq.finish_runqueue(True) (see cooker.py which does this).
>
> The return one could probably also be a finish_runqueue(False) call.

The finish_runqueue(False) works, but finish_runqueue(True) doesn't work
if there is no running task running(for example, when we use the
finish_runqueue(True) at the very beginning of the build), this is because
if there is no running taks, the function would do nothing. So I use:

rq.finish_runqueue(True)
return False

Then the build would always stop whether there is running tasks or not.

>>       def execute_runqueue(self):
>>           """
>>           Run the tasks in a queue prepared by rqdata.prepare()
>> @@ -946,7 +993,14 @@ class RunQueue:
>>                   self.rqexe = RunQueueExecuteScenequeue(self)
>>
>>           if self.state is runQueueSceneRun:
>> -            retval = self.rqexe.execute()
>> +            if self.dm and self.dm.enableMonitor:
>> +                dm_action = self.disk_monitor_action(self.dm.devDict)
>> +                if dm_action == 1:
>> +                    self.rqexe.finish()
>> +                else:
>> +                    retval = self.rqexe.execute()
>> +            else:
>> +                retval = self.rqexe.execute()
>>
>>           if self.state is runQueueRunInit:
>>               logger.info("Executing RunQueue Tasks")
>> @@ -954,10 +1008,17 @@ class RunQueue:
>>               self.state = runQueueRunning
>>
>>           if self.state is runQueueRunning:
>> -            retval = self.rqexe.execute()
>> +            if self.dm and self.dm.enableMonitor:
>> +                dm_action = self.disk_monitor_action(self.dm.devDict)
>> +                if dm_action == 1:
>> +                    self.rqexe.finish()
>> +                else:
>> +                    retval = self.rqexe.execute()
>> +            else:
>> +                retval = self.rqexe.execute()
>>
>>           if self.state is runQueueCleanUp:
>> -           self.rqexe.finish()
>> +            self.rqexe.finish()
>>
>>           if self.state is runQueueComplete or self.state is runQueueFailed:
>>               if self.rqexe.stats.failed:
>
>
> With the above changes, you can probably just put something like:
>
>         if self.state in [runQueueSceneRun, runQueueRunning, runQueueCleanUp]:
>              self.dm.check(self)
>

This would make the code more clearer, I have moved most of the code to
lib/bb/monitordisk.py as you suggested, but since the rq.finish_runqueue(True)
doesn't work when there is no running task, I still need check the return
status of the monitor, now the code is:

         if self.state in [runQueueSceneRun, runQueueRunning, runQueueCleanUp]:
             if self.dm.enableMonitor:
                 dm_ret = self.dm.dm_action()
                 if dm_ret == 1:
                     self.finish_runqueue(False)
                 elif dm_ret == 2:
                     self.finish_runqueue(True)
                     return False

// Robert

> in above if self.state is runQueueCleanUp in execute_runqueue() since
> calling the finish_runqueue() would change self.state
>
> Cheers,
>
> Richard
>
>
>
>