satya - 8/13/2016, 12:31:25 PM

Forked processes Don't get killed consistently

When users cancel a process from dashboard the processes don't get killed with consistency. This is a bit better in forked processes in the forked mode of a molecule. It is much worse in a non-forked molecule.

satya - 8/13/2016, 12:33:15 PM

Forked processes appear to hang....

When forked processes are waiting for something external like a SQL call or a mismatched table space in a database, they just seem to hang. No warning or heartbeat messages in log files. Would be nice to see a heartbeat message indicating what it is waiting on...

satya - 8/13/2016, 12:35:27 PM

When forked processes terminate the main process error reporting is not clear what happened

when a process fails due to low memory or out of memory or other errors like a file system error, we are not quickly able to debug what happened. For example we may get an unknown failure.

satya - 8/13/2016, 12:37:17 PM

So a process may have exhibit any of the following behaviors

Low memory
Out of memory
File reads on NFS
Status of a forked process
Map thrashing for memory
Seem to wait for ever for something
Not able to see log files midstream of a process

satya - 8/13/2016, 12:43:21 PM

It will be nice if I can see a frequent heart beat with

How many threads I have now
how much memory I am using and left
What I am waiting on
Where am I at

satya - 8/13/2016, 12:56:35 PM

what should I expect from a devleoper

Run in test in non-forked mode
Run it for 1000, 5000, 10K records
No flow controls
Give me a map of CPU 
Give me a map of Memory
(use java monitors)
Give me your batch size

satya - 8/13/2016, 12:59:58 PM

Also have them submit log files

this will allow to examine the process and get to analyze the timings of various shapes.