1.Capture process uses a lot of memory and creates lobcahe files (o.gprp-lob18324952.a & o.gprp-lob18324952.b in the VARDIR/state directory.
2.Capture Core Dumps and the core file is very large 4GB and it creates lobcahe files in the VARDIR/state directory.
In both scenarios you also see these messages in sp_ctrl at times.
sp_ctrl (2100)> RSP_MALLOC_FAILED at rmp.c:420; module rmp size = 14000 bytes
This occurs in environments where LOBS are replicated due to how shareplex handles rollbacks of lobs causing lob_ids not to be associated with a transaction thus causing a permanent memory leak in capture.
WORKAROUND:
1. Shutdown shareplex
2. Rename the cmap.a and cmap.b files & lob cache files in Vardir/State directory during a downtime .
3. Restart shareplex.
NOTE: If you perform the workaround at the time of replication then if there is a long running open transaction then you may be out of sync on this and on the LOB tables.
STATUS:
Fixed in Shareplex for Oracle version 5.3.4.
