Apparent memory leak in long running OpenVista processes

[Derek_]

Many long running OpenVista processes that serve TCP connections continually accumulate system memory. For example, the broker processes and HL interface processes gradually increase in memory usage over the course of days and weeks. The leak seems likely to be in low-level MUMPS routines adopted/modified for the GT.M Integration project.

[bhaskar (bhaskar)]

Which GT.M version? We did fix a memory leak in V5.3-004 - extract below from the release notes . Also, if long running processes don't close open devices, that can cause a memory leak

C9J01-003080 Memory leak plugged in pattern matching with alternation

GT.M now frees memory created when processing a pattern match operator containing alternations. In previous versions of GT.M, this same usage caused a small memory leak. (C9J01-003080)

[jeff.apple (jeff-apple)]

I am investigating version "V5.4-000A Linux x86".

The non-closed device had occurred to me, but I do not see any indication of that at this time. I will start looking into that possibility in greater detail.

[Derek_ (derek-name)]

Thanks.

The sites where this is being seen are running V5.4-000A.

[Derek_ (derek-name)]

Jeff and FIS have found the source of the memory leak to be in GT.M itself in the code handling ZSHOW "D" for sockets. And they have come up with strategies for a workaround. The OpenVista code where this happens is in ZISTCPS.

The fix in GT.M is expected in the upcoming release of GT.M. For the time until that version is released by Fidelity and packaged into this project, we will need to provide a workaround in the MUMPS code of this project.