MTXW and Package Deployments

Running 8.11 SP1 on iSeries with 8.95_N1 (same issue occurred on 8.95_J1).

When I deploy a full package to any environment the iSeries just hangs. All I get are MTXW and the system just pauses.

I've already tried the normal dddict, ddtext, glbtbl cleanips, SQLPKG clean ups and all that.

The real kicker is that ENDNET doesn't work after that and leaves some JDE jobs hanging. I then have to do a sequence of ENDJOB's to get the system down which doesn't always work. ENDJOBADN and ENDSBS do nothing since it's already trying to end. Long and short of it is that I'm often doing an IPL just to kill E1.

Anyone have anything similar? Any Suggestions?

Colin

Colin,

We had the same issue last week while deploying an update package (with 450 objects) and it sent a couple of the kernels into MTXW. We were told by Oracle that this is normal and to just sit tight while the deploy completes. 2 hours later, once the deploy completed on the iSeries, everything returned to normal and all MTXW kernels went back to normal.

Greg

Just to add to the weirdness if I change the number of call object kernals that autostart to 0 then all works well.

Here's my kernal definition for the call object kernal:

[JDENET_KERNEL_DEF6]
krnlName=CALL OBJECT KERNEL
dispatchDLLName=XMLCALLOBJ
dispatchDLLFunction=XMLCallObjectDispatch
maxNumberOfProcesses=30
numberOfAutoStartProcesses=10
;changed for testing
;numberOfAutoStartProcesses=0
singleThreadedMode=N
ThreadPoolSize=30
ThreadPoolSizeIncrement=10

Hi Colin;

I have seen this problem when deploying update or full packages to our
i5. It only happens to us if, during the deploy a UBE starts up. So to
solve it I always hold all of the E1 job queues before deploying any
package to the i5 and restart them as soon as the deploy is finished.
You are probably already doing that but I thought it at least worth
mentioning.



Also, as an alternative to using ENDSBS or IPLing the system, if you
have done an endjob *immed, you have to wait 10 minutes before you can
use the ENDJOBABN. When it happened to me I waited the ten minutes then
was able to use ENDJOBABN, and then the E1 subsystem ended and I was
able to restart it.

Re: RE: MTXW and Package Deployments

Thanks,

no jobs were being submitted but the package deploy just hangs.

I have done the 10 minute wait but I figure that I shouldn't have to wait 10 minutes and more so I shouldn't have to bounce to clear the MTXW.

...and the real killer is that if I change:

[JDENET_KERNEL_DEF6]
krnlName=CALL OBJECT KERNEL
numberOfAutoStartProcesses=10

to

[JDENET_KERNEL_DEF6]
krnlName=CALL OBJECT KERNEL
numberOfAutoStartProcesses=0

then everything works.

Even worse is that this was an "all of a sudden" event. No tools changes, na da.

I don't have solid answers but just so you know you are not alone.... We are running 8.10 with Tools release 8.95I2 on an ISeries. I have run into the same problem but not with every deploy. About one third of my deploys go into a MTXW for extended periods. We "think" it is related to runaway kernels. Even though I have been told this should NOT effect the deploy, we have noticed some kernels using up to 8 CPU % when the system should be quiet. No batch jobs are running and no one should be online. Our server deploys normally take about 5 to 10 minutes. So, waiting 2 hours for a MTXW to clear as previously suggested seems extreme. I usually bounce JDE after 30 minutes. We had to IPL one time because a kernel would not die. Some have suggested it is due to corrupt packages but after bouncing services, I have never had issues deploying the same package. I will be watching this thread with interest.