Problem with RT-Kernel?
Posted: 08 Jan 2018, 11:56
We are running Wildfly 10.1 on a CM3 Revolution Pi. A loop polls (via JNI) the inputs of the connected DI module every 50ms and notifies subscribed programs of the input state and counter values if wanted.
In my current use case the the RPi would be connected to a PLC and interpret the input states as production status (e.g. DI input 1 = production, input 2 = downtime with reason xy, ....) and communicates them via OPC to our MES Software. Additionally I am storing state changes in the local H2 db provided by wildfly (at least once per minute).
All this works fine - for a while at least. After a couple hours the java process seems to stop working. In fact the OS locks up almost completely.
I mocked the DI interface to exclude problems with hardware/driver/JNI. The problem still remains the same, even without accessing the hardware.
After some testing I saw that the jbd2 process (jbd2/mmcblk0p2-) goes to state "D" (interuptable sleep). After that every process that wants to write to disk also goes to this state. This only happens with the RPi kernel. With the last Jessi lite image and its kernel I cannot replicate this.
Does anybody have any ideas what I could do?
In my current use case the the RPi would be connected to a PLC and interpret the input states as production status (e.g. DI input 1 = production, input 2 = downtime with reason xy, ....) and communicates them via OPC to our MES Software. Additionally I am storing state changes in the local H2 db provided by wildfly (at least once per minute).
All this works fine - for a while at least. After a couple hours the java process seems to stop working. In fact the OS locks up almost completely.
I mocked the DI interface to exclude problems with hardware/driver/JNI. The problem still remains the same, even without accessing the hardware.
After some testing I saw that the jbd2 process (jbd2/mmcblk0p2-) goes to state "D" (interuptable sleep). After that every process that wants to write to disk also goes to this state. This only happens with the RPi kernel. With the last Jessi lite image and its kernel I cannot replicate this.
Does anybody have any ideas what I could do?