Today I had to restart the EOS managers following changes in order to use QuarkDB for the namespace and configuration. It happened twice that the manager crashed shortly after starting when processing the first requests. I have a log extract with the crash on one of the managers, I can send it on request.
Trying again to start did not produce the same error and the mgm started normally.
We have seen this ourselves a couple of times in production, indeed. I had a quick look and made an experimental fix for a potential buffer overflow inside TSealedEnvelope::decodeEnvelope, which will be available in the next release of libTokenAuthz. (1.2.3)
It seems to be unrelated to the namespace migration.