Running process after image save/stop/restart

On 2013-02-28, at 22:38, Louis LaBrunda <[hidden email]> wrote:

> If a process is running when an image is saved, is the process stopped at
> any point in particular?

The active process is stopped in the snapshot primitive (a.k.a. "image saving") and resumes after it on startup. All other processes are waiting on some semaphore anyway.

> If the saved image is started, is there any way the process can tell?

Not the process itself. But you surely keep the process in a class somewhere, and the class can arrange to get notified on startup by adding itself to the startup list. See addToStartUpList:.

> I'm running a process that keeps running for a look time.  It loops with a
> delay and in the loop gets the date and time.  If the date and time were
> obtained just before the save, they would be old at the time of the image
> restart and need to be refreshed.

Delays get adjusted after resuming from snapshot. So it should just work.

- Bert -

Hi Eliot,

Thanks for all the very valuable information.  I think I didn't ask my
question properly but with all the answers I was able to accomplish exactly
what I wanted.

I do have a question about one thing you said.  See below.

Lou

>On Thu, Feb 28, 2013 at 2:30 PM, Bert Freudenberg <[hidden email]>wrote:
>
>> On 2013-02-28, at 22:38, Louis LaBrunda <[hidden email]> wrote:
>>
>> > If a process is running when an image is saved, is the process stopped at
>> > any point in particular?
>>
>> The active process is stopped in the snapshot primitive (a.k.a. "image
>> saving") and resumes after it on startup. All other processes are waiting
>> on some semaphore anyway.
>>
>
>Nope.  No processes are stopped.  Processes are simply unable to run while
>in the snapshot.  The key to understanding this is understanding the
>scheduler (the Processor global, an instance of ProcessorSheduler).  It
>maintains an activeProcess and a set of runnable processes, implemented as
>an Array of lists of processes at the same priority.  A runnable process
>(one not sent suspend or not waiting on a semaphore) is either the
>activeProcess or on one of the lists.  This state is saved to the snapshot.
> At any time the scheduler's activeProcess is the first highest priority
>process in the set of runnable processes.  It will only be deposed as
>activeProcess when either another higher-priority process becomes runnable
>or it yields or suspends or waits on a semaphore.  If it suspends or waits
>on a semaphore it is removed from the set of runnable processes.  If it
>yields it gets sent to the back of its run-queue and the next process in
>that queue becomes the runnable process.  If it is the only runnable
>process at its priority yield is a noop.
>
>On loading the image the VM activates the activeProcess; the activeProcess
>will typically be in a method that has called the snapshot primitive and
>the system will continue, with the snapshot primitive answering true.  All
>the other processes in the run queues are runnable but, just as before the
>snapshot, can't be run because the activeProcess is still runnable.  But as
>soon as the activeProcess suspends, waits or yields, one of these processes
>may run.
>
>Several processes are terminated (terminate is suspend + run unwinds) on
>resuming the image.  This is done to inform the VM of additional state to
>make these processes run.  For example, the Delay process needs to tell the
>VM what the next delay expiry is on start-up.  The delay semaphore (the
>semaphore the VM signals when the current active delay expires) is saved in
>the image in the specialObjects array, so it persists across a snapshot,
>but the VM doesn't persist the current delay expiry time.
>
>So any long-running process which doesn't need to inform the VM of anything
>special at start-up will just keep truckin' along, and nothing need be
>done.  A snapshot really is like a fermata, a pause.  It is not some kind
>of shut-down.
>
>As a side-note what happens if one does
>
>     [Semaphore new wait] fork.
>    Processor activeProcess yield.
>
>?
>
>This creates a new process and adds it to the Processor's run queue.  Once
>the process has been sent fork the current process doesn't reference it
>since it has been popped from the current process's stack, and so the only
>reference to the new process is from one of the Processors' run queues, but
>it isn't running yet because the activeProcess (the one that sent fork) is

>running.  When the activeProcess yields the new process gets to run.  Once

>it has created the new semaphore it sends wait to it.  At this point it is
>removed from the Processor's run-queue and added to the semaphore's queue.
> So now there is a circular reference between the process and the semaphore
>and these are the only references to the process and the semaphore, and so
>both get garbage collected.

I think the forked process gets to run before the processor returns to the
yield line.  This is the way it works in VA Smalltalk and from looking at
the implementers of #fork, I think it is the way it works in Squeak.  I
could be wrong, if so please be kind, I have been following the news group
for a while but I'm new to playing with Squeak.

There were times when I wanted to create a new fork but didn't want it to
run until the method creating it finished.  So, I added #forkReady (and
friends) that would create the process but not run it right away.

>>
>> > If the saved image is started, is there any way the process can tell?
>>
>> Not the process itself. But you surely keep the process in a class
>> somewhere, and the class can arrange to get notified on startup by adding
>> itself to the startup list. See addToStartUpList:.
>>
>
>exactly. one could also e.g. poll OSProcess to get the process ID of the VM
>process and see if that changes, but that's a horrible hack, and once in a
>blue moon will fail (cuz process ids are not unique and get reused).
>
>> I'm running a process that keeps running for a look time.  It loops with a
>> > delay and in the loop gets the date and time.  If the date and time were
>> > obtained just before the save, they would be old at the time of the image
>> > restart and need to be refreshed.
>>
>>
>> Delays get adjusted after resuming from snapshot. So it should just work.
>>
>
>exactly.  the snapshot will look just like a long time on the run-queue
>whole the process is preempted by higher-priroity processes,
>
>
>> - Bert -
>>

-----------------------------------------------------------
Louis LaBrunda
Keystone Software Corp.
SkypeMe callto://PhotonDemon
mailto:[hidden email] http://www.Keystone-Software.com

On 2013-03-05, at 21:32, Eliot Miranda <[hidden email]> wrote:

On Thu, Feb 28, 2013 at 2:30 PM, Bert Freudenberg <[hidden email]> wrote:

On 2013-02-28, at 22:38, Louis LaBrunda <[hidden email]> wrote:

> If a process is running when an image is saved, is the process stopped at
> any point in particular?

The active process is stopped in the snapshot primitive (a.k.a. "image saving") and resumes after it on startup. All other processes are waiting on some semaphore anyway.

Nope.  No processes are stopped.  Processes are simply unable to run while in the snapshot.

Yes, I used sloppy language, "suspended" is loaded in this context. The active process does not go through a suspend/resume cycle, it's simply that time "stops" and later "continues", so to speak.

Excellent explanation, Eliot :)

- Bert -

 The key to understanding this is understanding the scheduler (the Processor global, an instance of ProcessorSheduler).  It maintains an activeProcess and a set of runnable processes, implemented as an Array of lists of processes at the same priority.  A runnable process (one not sent suspend or not waiting on a semaphore) is either the activeProcess or on one of the lists.  This state is saved to the snapshot.  At any time the scheduler's activeProcess is the first highest priority process in the set of runnable processes.  It will only be deposed as activeProcess when either another higher-priority process becomes runnable or it yields or suspends or waits on a semaphore.  If it suspends or waits on a semaphore it is removed from the set of runnable processes.  If it yields it gets sent to the back of its run-queue and the next process in that queue becomes the runnable process.  If it is the only runnable process at its priority yield is a noop.

On loading the image the VM activates the activeProcess; the activeProcess will typically be in a method that has called the snapshot primitive and the system will continue, with the snapshot primitive answering true.  All the other processes in the run queues are runnable but, just as before the snapshot, can't be run because the activeProcess is still runnable.  But as soon as the activeProcess suspends, waits or yields, one of these processes may run.

Several processes are terminated (terminate is suspend + run unwinds) on resuming the image.  This is done to inform the VM of additional state to make these processes run.  For example, the Delay process needs to tell the VM what the next delay expiry is on start-up.  The delay semaphore (the semaphore the VM signals when the current active delay expires) is saved in the image in the specialObjects array, so it persists across a snapshot, but the VM doesn't persist the current delay expiry time.

So any long-running process which doesn't need to inform the VM of anything special at start-up will just keep truckin' along, and nothing need be done.  A snapshot really is like a fermata, a pause.  It is not some kind of shut-down.

As a side-note what happens if one does

     [Semaphore new wait] fork.

    Processor activeProcess yield.

?

This creates a new process and adds it to the Processor's run queue.  Once the process has been sent fork the current process doesn't reference it since it has been popped from the current process's stack, and so the only reference to the new process is from one of the Processors' run queues, but it isn't running yet because the activeProcess (the one that sent fork) is running.  Wen the activeProcess yields the new process gets to run.  Once it has created the new semaphore it sends wait to it.  At this point it is removed from the Processor's run-queue and added to the semaphore's queue.  So now there is a circular reference between the process and the semaphore and these are the only references to the process and the semaphore, and so both get garbage collected.

 

> If the saved image is started, is there any way the process can tell?

Not the process itself. But you surely keep the process in a class somewhere, and the class can arrange to get notified on startup by adding itself to the startup list. See addToStartUpList:.

exactly. one could also e.g. poll OSProcess to get the process ID of the VM process and see if that changes, but that's a horrible hack, and once in a blue moon will fail (cuz process ids are not unique and get reused).

> I'm running a process that keeps running for a look time.  It loops with a
> delay and in the loop gets the date and time.  If the date and time were
> obtained just before the save, they would be old at the time of the image
> restart and need to be refreshed.

Delays get adjusted after resuming from snapshot. So it should just work.

exactly.  the snapshot will look just like a long time on the run-queue whole the process is preempted by higher-priroity processes,

 

- Bert -

--
HTH,

Eliot

I will bow to your knowledge of Squeak but take a look at the code below
from VA Smalltalk.

Pay special attention to #resume: which looks to me like it puts the new
process at the top of the queue (there are 7 queues, one for each priority)
and puts the current (activeProcess) at the bottom of the same queue and
then switches to the highestPriorityProcess.  Which can be another process
or the new process just put in front of the current process.

Don't ask my why it works this way.  As I said, I added #forkReady which it
seems works more like Squeak.  Most of the time I think it doesn't matter
but sometimes it does.

Lou

fork
        "Create a new Process which is scheduled by the
         ProcessScheduler. The new process executes the receiver by
         sending it the message value. The new process is created
         with the same prority as the activeProcess. Answer the
         receiver."

        ^self newProcess resume

resume
        "Tell the process scheduler to add the process to the ready
         to run queue."

        self isResumable ifFalse: [^self error: (NlsCatKRN indexedMsg: 2)].
"$NLS$ process cannot be resumed"
        Processor resume: self.
        ^ self

resume: aProcess

        | state |
        state := self enableAsyncMessages: false.
        (aProcess == self activeProcess or: [aProcess isDead]) ifTrue: [
                self enableAsyncMessages: state.
                ^self "Resuming a dead process or the active process is a nop."].

        aProcess isRunable ifTrue: [
                (aProcess processState: ##ready) queue addFirst: aProcess].

        (self activeProcess processState: ##ready) queue addLast: self
activeProcess.
        self activeProcess switchTo: self highestPriorityProcess.
        self enableAsyncMessages: state








On Tue, 5 Mar 2013 13:46:13 -0800, Eliot Miranda <[hidden email]>
wrote:

>On Tue, Mar 5, 2013 at 1:36 PM, Louis LaBrunda <[hidden email]>wrote:
>
>> Hi Eliot,
>>
>> Thanks for all the very valuable information.  I think I didn't ask my
>> question properly but with all the answers I was able to accomplish exactly
>> what I wanted.
>>
>> I do have a question about one thing you said.  See below.
>>
>> Lou
>>
>> >On Thu, Feb 28, 2013 at 2:30 PM, Bert Freudenberg <[hidden email]
>> >wrote:
>> >
>> >> On 2013-02-28, at 22:38, Louis LaBrunda <[hidden email]>
>> wrote:
>> >>
>> >> > If a process is running when an image is saved, is the process
>> stopped at
>> >> > any point in particular?
>> >>
>> >> The active process is stopped in the snapshot primitive (a.k.a. "image
>> >> saving") and resumes after it on startup. All other processes are
>> waiting
>> >> on some semaphore anyway.
>> >>
>> >
>> >Nope.  No processes are stopped.  Processes are simply unable to run while
>> >in the snapshot.  The key to understanding this is understanding the
>> >scheduler (the Processor global, an instance of ProcessorSheduler).  It
>> >maintains an activeProcess and a set of runnable processes, implemented as
>> >an Array of lists of processes at the same priority.  A runnable process
>> >(one not sent suspend or not waiting on a semaphore) is either the
>> >activeProcess or on one of the lists.  This state is saved to the
>> snapshot.
>> > At any time the scheduler's activeProcess is the first highest priority
>> >process in the set of runnable processes.  It will only be deposed as
>> >activeProcess when either another higher-priority process becomes runnable
>> >or it yields or suspends or waits on a semaphore.  If it suspends or waits
>> >on a semaphore it is removed from the set of runnable processes.  If it
>> >yields it gets sent to the back of its run-queue and the next process in
>> >that queue becomes the runnable process.  If it is the only runnable
>> >process at its priority yield is a noop.
>> >
>> >On loading the image the VM activates the activeProcess; the activeProcess
>> >will typically be in a method that has called the snapshot primitive and
>> >the system will continue, with the snapshot primitive answering true.  All
>> >the other processes in the run queues are runnable but, just as before the
>> >snapshot, can't be run because the activeProcess is still runnable.  But
>> as
>> >soon as the activeProcess suspends, waits or yields, one of these
>> processes
>> >may run.
>> >
>> >Several processes are terminated (terminate is suspend + run unwinds) on
>> >resuming the image.  This is done to inform the VM of additional state to
>> >make these processes run.  For example, the Delay process needs to tell
>> the
>> >VM what the next delay expiry is on start-up.  The delay semaphore (the
>> >semaphore the VM signals when the current active delay expires) is saved
>> in
>> >the image in the specialObjects array, so it persists across a snapshot,
>> >but the VM doesn't persist the current delay expiry time.
>> >
>> >So any long-running process which doesn't need to inform the VM of
>> anything
>> >special at start-up will just keep truckin' along, and nothing need be
>> >done.  A snapshot really is like a fermata, a pause.  It is not some kind
>> >of shut-down.
>> >
>> >As a side-note what happens if one does
>> >
>> >     [Semaphore new wait] fork.
>> >    Processor activeProcess yield.
>> >
>> >?
>> >
>> >This creates a new process and adds it to the Processor's run queue.  Once
>> >the process has been sent fork the current process doesn't reference it
>> >since it has been popped from the current process's stack, and so the only
>> >reference to the new process is from one of the Processors' run queues,
>> but
>> >it isn't running yet because the activeProcess (the one that sent fork) is
>> >running.  When the activeProcess yields the new process gets to run.  Once
>> >it has created the new semaphore it sends wait to it.  At this point it is
>> >removed from the Processor's run-queue and added to the semaphore's queue.
>> > So now there is a circular reference between the process and the
>> semaphore
>> >and these are the only references to the process and the semaphore, and so
>> >both get garbage collected.
>>
>> I think the forked process gets to run before the processor returns to the
>> yield line.  This is the way it works in VA Smalltalk and from looking at
>> the implementers of #fork, I think it is the way it works in Squeak.  I
>> could be wrong, if so please be kind, I have been following the news group
>> for a while but I'm new to playing with Squeak.
>>
>
>No.  In Smalltalk-80 (VisualWorks, Squeak etc) fork creates a new process
>with the same priority as the current process and then resumes it (resume
>is the primitive that adds the process to the run queue).  So the new
>process is effectively behind the activeProcess in the run-queue.  e.g.
>
>| who |
>who := #me.
>[who := #him. Semaphore new wait] fork.
>who
>=> #me
>
>
>| who |
>who := #me.
>[who := #him. Semaphore new wait] fork.
>Processor yield.
>who
>=> #him
>
>I can't speak for VA, but I doubt you;re right.  I expect VA to have the
>same behaviour as the above.
>
>
>
>> There were times when I wanted to create a new fork but didn't want it to
>> run until the method creating it finished.  So, I added #forkReady (and
>> friends) that would create the process but not run it right away.
>>
>
>there's also forkAt: Processor activePriority - 1, or
>
>| gate result |
>gate := Semaphore new.
>[gate wait.
> self doStuff] fork.
>result := self getResult.
>gate signal.
>^result
>
>
>> >>
>> >> > If the saved image is started, is there any way the process can tell?
>> >>
>> >> Not the process itself. But you surely keep the process in a class
>> >> somewhere, and the class can arrange to get notified on startup by
>> adding
>> >> itself to the startup list. See addToStartUpList:.
>> >>
>> >
>> >exactly. one could also e.g. poll OSProcess to get the process ID of the
>> VM
>> >process and see if that changes, but that's a horrible hack, and once in a
>> >blue moon will fail (cuz process ids are not unique and get reused).
>> >
>> >> I'm running a process that keeps running for a look time.  It loops
>> with a
>> >> > delay and in the loop gets the date and time.  If the date and time
>> were
>> >> > obtained just before the save, they would be old at the time of the
>> image
>> >> > restart and need to be refreshed.
>> >>
>> >>
>> >> Delays get adjusted after resuming from snapshot. So it should just
>> work.
>> >>
>> >
>> >exactly.  the snapshot will look just like a long time on the run-queue
>> >whole the process is preempted by higher-priroity processes,
>> >
>> >
>> >> - Bert -
>> >>
>> -----------------------------------------------------------
>> Louis LaBrunda
>> Keystone Software Corp.
>> SkypeMe callto://PhotonDemon
>> mailto:[hidden email] http://www.Keystone-Software.com
>>
>>
>>
-----------------------------------------------------------
Louis LaBrunda
Keystone Software Corp.
SkypeMe callto://PhotonDemon
mailto:[hidden email] http://www.Keystone-Software.com