Debug build doesn't produce a core.

No tests which test traffic pass until Fragments in fragments reassembly. After that, each test fails with IOError: [Errno 2] Connect failed

There are also some API failures:
UnexpectedApiReturnValueError: API call failed, expected 0 return value instead of -127 in sw_interface_add_del_address_reply(_0=89, context=2316, retval=-127)

For reference, here's what -127 is:
_(DUPLICATE_IF_ADDRESS, -127, "Address already present on another interface")

After all tests are executed the framework hangs and there is one vpp still running, the one from TestFIFReassembly. A bit of gdb poking reveals that the vpp actually didn't halt, but is doing something with memory.

I've been able to identify two offending patches:
https://gerrit.fd.io/r/#/c/15106/
https://gerrit.fd.io/r/#/c/15534/

Everything was passing before the first patch (all tests passed). The first patch introduced this behavior:
==============================================================================
Dummy test case used to check if VPP is able to start
==============================================================================
connect: No such file or directory
Traceback (most recent call last):
File "sanity_run_vpp.py", line 21, in <module>
tc.setUpClass()
File "/home/jlinkes/vpp/test/framework.py", line 423, in setUpClass
cls.statistics = VPPStats(socketname=cls.stats_sock)
File "build/bdist.linux-aarch64/egg/vpp_papi/vpp_stats.py", line 120, in _init_
IOError
*******************************************************************
* Sanity check failed, cannot run vpp
*******************************************************************

Only after the second patch it's possible to reproduce the current behavior.