BTW there is cancel to cancel stopBroadcast() if flow is restored the timeout seems to be insufficient ... :(
On Thu, 11 Feb 2021 at 08:39, Maxim Solodovnik <[email protected]> wrote: > > > On Thu, 11 Feb 2021 at 02:26, [email protected] <[email protected]> > wrote: > >> According to my observations: if stream is NOT_FLOWING it will never flow >> again :( >> this is why the connection is dropped >> >> That might be true for MediaType VIDEO but NOT for MediaType *AUDIO* >> >> In the log file (I've added streamIds to the logs) you can see NOT_FLOWING >> but later again you can see FLOWING events. For the *same* streamId >> >> Example 1 >> [34mINFO [0;39m 02-10 08:11:38.220 [36mo.a.o.c.r.KStream:160 >> [ventExec-e2-t48] [0;39m - Media Flow STATE :: *NOT_FLOWING*, type >> MediaFlowOutStateChange, evt *AUDIO*, sid >> *d6c5db69-0d30-4ba6-b863-aa3c297e17ac*, uid >> 9ad475f3-e994-4850-97a3-3e3717bf755a >> [34mINFO [0;39m 02-10 08:11:47.022 [36mo.a.o.c.r.KStream:160 >> [ventExec-e2-t73] [0;39m - Media Flow STATE :: *FLOWING*, type >> MediaFlowOutStateChange, evt *AUDIO*, sid >> *d6c5db69-0d30-4ba6-b863-aa3c297e17ac*, uid >> 9ad475f3-e994-4850-97a3-3e3717bf755a >> >> Example 2 >> [34mINFO [0;39m 02-10 08:11:52.066 [36mo.a.o.c.r.KStream:160 >> [ventExec-e2-t65] [0;39m - Media Flow STATE :: *NOT_FLOWING*, type >> MediaFlowOutStateChange, evt *AUDIO*, sid >> *b8e9f1a6-d29f-482c-8faf-4ac9b9939527*, uid >> eb6a7aa7-b913-4d51-a34f-ab1ac684f27d >> [34mINFO [0;39m 02-10 08:12:03.501 [36mo.a.o.c.r.KStream:160 >> [ventExec-e2-t51] [0;39m - Media Flow STATE :: *FLOWING*, type >> MediaFlowOutStateChange, evt *AUDIO*, sid >> *b8e9f1a6-d29f-482c-8faf-4ac9b9939527*, uid >> eb6a7aa7-b913-4d51-a34f-ab1ac684f27d >> >> Example 3 >> [34mINFO [0;39m 02-10 08:12:16.099 [36mo.a.o.c.r.KStream:160 >> [ventExec-e2-t40] [0;39m - Media Flow STATE :: *NOT_FLOWING*, type >> MediaFlowOutStateChange, evt *AUDIO*, sid >> *46420fab-fcaa-493f-9283-1b2fbf9f3170*, uid >> b996144c-9ab1-43d5-8500-da359446ad13 >> [34mINFO [0;39m 02-10 08:12:19.305 [36mo.a.o.c.r.KStream:160 >> [ventExec-e2-t40] [0;39m - Media Flow STATE :: *FLOWING*, type >> MediaFlowOutStateChange, evt *AUDIO*, sid >> *46420fab-fcaa-493f-9283-1b2fbf9f3170*, uid >> b996144c-9ab1-43d5-8500-da359446ad13 >> >> I have copied the full log of all MediaFlow State events when doing the >> 140 >> users test here: https://pastebin.com/ZEmtZVyH >> >> => You can find another 20 examples of FLOWING after NOT_FLOWING for type >> AUDIO. You can _never_ find it for VIDEO. But type AUDIO can indeed start >> flowing again! >> >> I understand that this might be a challenge to detect stop events if we >> can't use this one. But just to be clear, I'm talking about type AUDIO. >> There must be a way to filter them out to not kill the stream in those >> cases. >> > > Great! > Thanks for the clarifications! > I'll add this check to the master ASAP! > > >> >> Thanks >> Sebastian >> >> Sebastian Wagner >> Director Arrakeen Solutions, OM-Hosting.com >> http://arrakeen-solutions.co.nz/ >> https://om-hosting.com - Cloud & Server Hosting for HTML5 >> Video-Conferencing OpenMeetings >> < >> https://www.youracclaim.com/badges/da4e8828-743d-4968-af6f-49033f10d60a/public_url >> > >> < >> https://www.youracclaim.com/badges/b7e709c6-aa87-4b02-9faf-099038475e36/public_url >> > >> >> >> On Wed, 10 Feb 2021 at 23:53, Maxim Solodovnik <[email protected]> >> wrote: >> >> > On Wed, 10 Feb 2021 at 16:22, [email protected] < >> [email protected] >> > > >> > wrote: >> > >> > > I was able to fix the missing Video Pods issue. And I think it was a >> > > successful test with 140 users. >> > > >> > > See: >> > > >> > > >> > >> https://cwiki.apache.org/confluence/display/OPENMEETINGS/OpenMeetings+140+users+test+10-02-2021+with+256crypt+and+FLOW_STATE+fix+for+not+terminating+AUDIO >> > > >> > > Fixes in the branch I was testing: >> > > >> > > - The previously described fix to reduce the N factor in SCrypt >> > > calculation >> > > - The previous fix for adding some DB INdex on Address.email column >> > > >> > > And also in this one: Fix for not calling "stopBroadcast" when >> > > MediaFlowState.NOT_FLOWING is of type AUDIO. >> > > => I discovered in previous tests that there are almost exactly the >> > amount >> > > of video pods missing compared to the log event: >> > > [34mINFO [0;39m 02-10 08:11:38.220 [36mo.a.o.c.r.KStream:160 >> > > [ventExec-e2-t48] [0;39m - Media Flow STATE :: NOT_FLOWING, type >> > > MediaFlowOutStateChange, evt *AUDIO*, sid >> > > d6c5db69-0d30-4ba6-b863-aa3c297e17ac, uid >> > > 9ad475f3-e994-4850-97a3-3e3717bf755a >> > > >> > > I think AUDIO can stop flowing! there is no need to kill the stream in >> > this >> > > scenario. >> > > >> > > To test and verify I updated my performance investigating branch to >> > modify >> > > and add && MediaType.VIDEO == evt.getMediaType(): before >> > > >> > > >> > >> https://github.com/apache/openmeetings/blob/feature/OPENMEETINGS-2567-investigate-performance-monitoring/openmeetings-core/src/main/java/org/apache/openmeetings/core/remote/KStream.java#L162 >> > > >> > > And I repeated the 140 user test and it's fine now! (well a lot >> better!) >> > > >> > >> > This code was added because clients thought everything is OK while other >> > people in the room were not seeing/hearing >> > And such misunderstanding is much worse than just reloading :( >> > I believe the change like this need to be tested extremely carefully >> > >> > According to my observations: if stream is NOT_FLOWING it will never >> flow >> > again :( >> > this is why the connection is dropped >> > >> > >> > > During this test there were almost 100 "Media Flow STATE :: >> NOT_FLOWING, >> > > type AUDIO" events BUT as you can see in the screenshots most video >> pods >> > > are fine: >> > > >> > > >> > >> https://cwiki.apache.org/confluence/display/OPENMEETINGS/OpenMeetings+140+users+test+10-02-2021+with+256crypt+and+FLOW_STATE+fix+for+not+terminating+AUDIO#OpenMeetings140userstest10022021with256cryptandFLOW_STATEfixfornotterminatingAUDIO-ScreenshotsfromeachroomwithVideoPods >> > > >> > > There are other issues :) There is a low percentage of video pods >> hanging >> > > (see screenshots on some pods). But previously those pods were >> completely >> > > missing and the ratio of pods missing was 50-70%!! >> > > => So I still call this a success :) Since it's a massive improvement >> > from >> > > the results before. >> > > >> > > I will raise a separated PR for discussing lowering the SCrypt value >> (1) >> > > and how to address the issue around MediaState.NOT_FLOWING and type >> AUDIO >> > > to not close the broadCast in those cases (2). >> > > >> > > Unfortunate NOT_FLOWING for type AUDIO, I'm not sure but we may not be >> > able >> > > to use this. I can also remember from the past that gaps in Audio >> streams >> > > can happen in a normal streaming scenario. Cause streaming tries to >> save >> > > bandwidth. So NOT_FLOWING for Audio may actually not mean anything. >> > > >> > > Thanks, >> > > Seb >> > > >> > > Sebastian Wagner >> > > Director Arrakeen Solutions, OM-Hosting.com >> > > http://arrakeen-solutions.co.nz/ >> > > https://om-hosting.com - Cloud & Server Hosting for HTML5 >> > > Video-Conferencing OpenMeetings >> > > < >> > > >> > >> https://www.youracclaim.com/badges/da4e8828-743d-4968-af6f-49033f10d60a/public_url >> > > > >> > > < >> > > >> > >> https://www.youracclaim.com/badges/b7e709c6-aa87-4b02-9faf-099038475e36/public_url >> > > > >> > > >> > >> > >> > -- >> > Best regards, >> > Maxim >> > >> > > > -- > Best regards, > Maxim > -- Best regards, Maxim
