Solution for flushing (and replacing) cached external 64 bit routines (db2fmp processes)

Solution for flushing (and replacing) cached external 64 bit routines (db2fmp processes)

We are running in our online production environment thousands external db2 routines per minute. Because of its performance we are forced to cache these routines by setting dbm cfg keepfenced to yes.

The database must be available 24/7. Of course, we have at least once per day the urgent necessity for hotfixes (new versions of self written cobol modules) without stopping the database.

So, we need a solution which flushes pooled db2fmp processes (and running db2fmp processes after they return to pool) similar to the solution which exists for db2fmp32 processes.

Solution for flushing (and replacing) cached external 64 bit routines (db2fmp processes)

We are running in our online production environment thousands external db2 routines per minute. Because of its performance we are forced to cache these routines by setting dbm cfg keepfenced to yes.

The database must be available 24/7. Of course, we have at least once per day the urgent necessity for hotfixes (new versions of self written cobol modules) without stopping the database.

So, we need a solution which flushes pooled db2fmp processes (and running db2fmp processes after they return to pool) similar to the solution which exists for db2fmp32 processes.

IBM implemented this feature for 32-bit routines as result of Case TS005866848.

IBM implementation details for 32-bit routines (as far as I understand):

1. DB2SET DB2_FMP_CLEAN_DELAY=90000

2. UPDATE DBM CFG USING FENCED_POOL 0 IMMEDIATE

3. UPDATE DBM CFG USING FENCED_POOL -1 AUTOMATIC IMMEDIATE

4. AFTER A FEW MINUTES: DB2SET DB2_FMP_CLEAN_DELAY=

By setting FENCED_POOL to 0, the states of all db2fmp processes are set to 0x02000010.

When these pooled processes are reused, they are terminated first and then new processes are started.

A second requirement:

Sometimes, we have problems with looping db2fmp processes.

Therefore, it should also be possible to flush one single db2fmp process by setting the state to 0x02000010.

Same AHA request for 32-bit routines exists: https://ibm-data-and-ai.ideas.ibm.com/ideas/DB24LUW-I-1365