~icecast/icecast/kh : changes

~icecast/icecast/kh » Changes from revision 877

From Revision 877 to 858

Rev	Summary	Authors	Tags	Date
877	Merge pull request #257 from trilliot/add-icy-metadata-cors Merge pull request #257 from trilliot/add-icy-metadata-cors add Icy-MetaData to Access-Control-Allow-Headers	Karl Heyes		4 years ago
876	last cleanups for versioning/typos last cleanups for versioning/typos	Karl Heyes		4 years ago
875	push a kh13 release out push a kh13 release out	Karl Heyes		4 years ago
874	fix possible null deref from previous commit and optimise th... fix possible null deref from previous commit and optimise the no lag case when joining queue	Karl Heyes		4 years ago
873	allow for seconds to be specified in burst, queue and min qu... allow for seconds to be specified in burst, queue and min queue settings we still allow for the numbers as before but if you want to use seconds then specify like <burst-size>5s</burst-size> the awkward bit has been the calculation of bitrate assessment. As there is a variable element to it (eg relays getting a burst) then the calculation is re-assessed every few seconds. This should allow for the default to be specified in secs for any stream, eg 30 secs for queue and may be 6 seconds for burst and that would apply to 32k or 128k streams.	Karl Heyes		4 years ago
872	memory leak and corruption fix memory leak and corruption fix intro content from auth and small YP leak fixed. race condition leading to use after free bug with listener connection getting closed when queue shrinkage occurs other small internal cleanups	Karl Heyes		4 years ago
871	missing on-demand setting in master block. missing on-demand setting in master block. This is just like an alias for the relays-on-demand setting. If the multiple master facility gets done then it can be used specifically for that master.	Karl Heyes		4 years ago
870	fix race case, adjust locking area when listeners move or dr... fix race case, adjust locking area when listeners move or drop off. In some cases, when moving to another worker or leaving a stream, there is a point where a source is referred to but not locked as it should be. By changing the locking constraints a lock/data is being maintained even if the source is to be freed up.	Karl Heyes		4 years ago
869	add some tolerance when pruning queue add some tolerance when pruning queue normally it should not matter but there is a race case between source client pruning and a listener detaching from queue when the lock is re-acquired for updating.	Karl Heyes		4 years ago
868	prevent incorrectly defined mime types from breaking relays prevent incorrectly defined mime types from breaking relays This is not usually a problem, but if one is define and it is not a recognised type then some initialisation of relays can be missed out on retrying. Initially when relays a created they are setup and work fine, but on retry, the structures and settings are not cleared out as type checking for listeners is maintained but initialisation needs to be done for certain things like metadata inserts. This mismatch was tripping a bug here causing metadata alignment problems. Now we verify the specified type before overriding.	Karl Heyes		4 years ago
867	enforce a / on local-mount for relays. enforce a / on local-mount for relays. if that is missing then it can lead to a mismatch between relay and source and mountpoint names. possible crash or hang on reload.	Karl Heyes		4 years ago
866	fix silly return value, largely ignored for now fix silly return value, largely ignored for now	Karl Heyes		4 years ago
865	avoid generic reader reschedule if block has been added. and... avoid generic reader reschedule if block has been added. and clean up silly log message silly log message was from the intro file load when there isn't one to load the skip setting was adversely affecting the reader part of certain high bandwidth streams. These are typically Mb/s. The rescheduling ended up increasing in certain cases to a point where lowering it would take it so long that the source would drop. This has shown when the source does not send anything straight away then starts blasting the content	Karl Heyes		4 years ago
864	enforce a range check in the queue sizing. enforce a range check in the queue sizing. seems like the assessed range could get messed up because clients could drop out and any time and leave the targeting too short. You could also get a lagging client also stretching the queue but that is less of an issue as that tends to resolve on a later interation. The former though could cause excessive purging.	Karl Heyes		4 years ago
863	closed socket and block sizing update changes closed socket and block sizing update changes the read test for closed proxy socket was far too frequent on high bitrates, so increase the frequency of them and make sure the nothing to do case also follows that same routine. make the automatic block size increase at a lower threshold, this should increase the client reschedule time slightly. Should only apply for high bitrates.	Karl Heyes		4 years ago
862	the internal incoming rate setting not properly set with rel... the internal incoming rate setting not properly set with relays the stats are ok as those are independently assessed but there are certain checks based on the internal setting that reference this and was only being done for source clients.	Karl Heyes		4 years ago
861	make adjustments to timings on stall and queue purge assessm... make adjustments to timings on stall and queue purge assessment The listeners need to respond to their positioning for accurate queue sizing, but with large retries it wa not happening which could lead to a fault if when listeners eventually moved off the queue. Here we exptend the assessment time a little, reduce the frequency it is done and reduced some of the max time frame the listener is tried again.	Karl Heyes		5 years ago
860	update for kh12 update for kh12	Karl Heyes	icecast-2.4.0-kh12	5 years ago
859	useful for core dump analysis when version info is not so cl... useful for core dump analysis when version info is not so clear	Karl Heyes		5 years ago
858	slight rework for the global max listeners assessment slight rework for the global max listeners assessment we still use the global listeners count, but the stats are updated on the regular update, no inc/dec routines are used now. There was a bug in the way from fallback file handling was done in this required which could affect the global max listeners chaeck. This should be resolved now	Karl Heyes		5 years ago

Older »