[agw] [Sctpd] [bug-fix v1.3] Re-add PartOf dependency between MME and sctpd

[ssanadhya]

Signed-off-by: Shruti Sanadhya ssanadhya@fb.com

Summary

After an MME crash, Sctpd goes into a crash loop while trying to acquire the socket. This change fixes the crash loop by (re-)adding the systemd dependency between Sctpd and MME service.

Note: This systemd dependency will cause Sctpd to restart even when stateless flag is set to true, hence commenting those out from the sanity test suite, while we work on a fix.

Test Plan

As @ulaskozat identified, the issue can be reproduced by running the test_attach_detach.py test from test VM and then kill the MME service on gateway with kill -9. Sctpd loops on

Dec 26 06:25:41 magma-dev sctpd[11972]: I1226 06:25:41.293893 12119 sctpd_downlink_impl.cpp:37] SctpdDownlinkImpl::Init starting
Dec 26 06:25:41 magma-dev sctpd[11972]: I1226 06:25:41.293928 12119 sctpd_downlink_impl.cpp:53] SctpdDownlinkImpl::Init creating new socket and list
Dec 26 06:25:41 magma-dev sctpd[11972]: I1226 06:25:41.293962 12119 util.cpp:58] sctp_bindx error (98): Address already in use

After applying this change, followed by:

• sudo systemctl daemon-reload
• sudo service magma@* stop
• sudo service magma@magmad start

Sending a kill -9 to MME does not lead to a crash loop.

Tested sanity with S1ap tests

Additional Information

• This change is backwards-breaking
  This needs an update of the Sctpd package