tree 1d8344ebaea933a8a3ca6b951b24fa3b2b764953
parent 996d4d7468f0857f7c192c27eaa23b15ba568ba0
author Primiano Tucci <primiano@google.com> 1522368902 +0100
committer Primiano Tucci <primiano@google.com> 1522411993 +0000

Reset ProbesProducer's state if losing service connection

Fixes a lifetime bug in traced_probes. The bug is the following:
when we lose the connection to the service we reset the IPC
|endpoint_| and re-attempt the connection. However the |endpoint_|
is the object that owns the shared memory buffer, on which the
various TraceWriter instances hold onto. We cannot destroy the
|endpoint_| without guaranteeing that all the TraceWriter(s) have
been destroyed as well.
This CL achieves this by simply invoking the destructor (which in
turn will destroy all the SinkDelegates and the various controllers)
and then back the constructor, starting with a brand new instance.
This, of course, relies on destruction order to be correct. But at
least doesn't introduce a secondary teardown path.

Bug: 77282805
Test: see repro steps of b/77282805.
Change-Id: Ieec7ab934c796222fea15365ee7c98d2db75cd6d