RTC::Consumer::GetStats() | failed assertion `it != this->mapProfileRtpStream.end()': effective profile does not map to a rtp stream

[ibc]

INFO:Participant#db6bivay consumer "effectiveprofilechange" event [profile:high]
INFO:Participant#db6bivay consumer "effectiveprofilechange" event [profile:none]
INFO:Participant#db6bivay consumer "effectiveprofilechange" event [profile:medium]
INFO:Participant#jxull4dn consumer "effectiveprofilechange" event [profile:high]
mediasoup-worker's stderr: ABORT[id:zxvppdey#1] RTC::Consumer::GetStats() | failed assertion `it != this->mapProfileRtpStream.end()': effective profile does not map to a rtp stream
mediasoup:ERROR:Worker child process exited [id:zxvppdey#1, code:null, signal:SIGABRT]

It happened once having stats retrieval enabled.

[ibc]

It seems that the same may crash on Consumer::GetLossPercentage().

[ibc]

It seems to me that effectiveProfile sometimes may point to targetProfile while such a profile does not exist in the map.

[jmillan]

It seems to me that effectiveProfile sometimes may point to targetProfile while such a profile does not exist in the map.

your suspicion is correct :-)

[ibc]

It has crashed again in 2.0.4:

mediasoup-worker's stderr: ABORT[id:yoydtkfb#1] RTC::Consumer::GetStats() | failed assertion `it != this->mapProfileRtpStream.end()': effective profile does not map to a rtp stream

[ibc]

It happens very often by just running:

CONSUMER.on('stats', (stats) => console.warn(JSON.stringify(stats, null, '  ')));
CONSUMER.enableStats();

AFAIS mostly when high stream dies.

[ibc]

AFAIS this commit e28738f fixed nothing. RecalculateTargetProfile() does not call to SetEffectiveProfile() at all.

// Target profile removed. Recalculate.
else if (this->targetProfile == profile)
{
	RecalculateTargetProfile();
}

It may perfectly happen that both targetProfile and effectiveProfile are the same, and also that it the removed profile. We are not updating effectiveProfileto NONE in that case so things crash.

[ibc]

Proposal: we need a test unit for the Consumer that provides source profiles, remove them, and checks that the effective and target profiles are always the proper ones.

[ibc]

a961e11 seems to fix the crash, and I would also say that I don't see frozen video anymore (#164) but I cannot confirm it at 100%.

Said that, there is something I don't like in the Consumer::RemoveProfile() method:

https://github.com/versatica/mediasoup/blob/master/worker/src/RTC/Consumer.cpp#L356

Why are we setting there the new targetProfile rather than relying on the RecalculateTargetProfile() method? IMHO we should rely on a single method to set the target profile (once we enable the effective one and add/remove available profiles).

[ibc]

Theoretically fixed in 2.0.5