Repro (one way, internal only due to found while investigating a b/):

(not a 100% repro since this is a race between operations on 2 sourcebuffers):

1) Make sure https://chromium-review.googlesource.com/c/552824/ is applied, otherwise this later DCHECK failure won't be reached 
2) With a debug build of Chrome, repro b/62450277 using its info in its comment 33, and with verbose MSE vmodule logging.

Successful repro indicates something like:

:0628/130342.048676:ERROR:render_media_log.cc(30)] MediaEvent: MEDIA_ERROR_LOG_ENTRY {"error":"Append: stream parsing failed. Data size=43411 append_window_start=0 append_window_end=inf"}
[1:1:0628/130342.072895:FATAL:chunk_demuxer.cc(1054)] Check failed: state_ == ENDED (4 vs. 3)

Stack shows this DCHECK failure is due to a new SourceBuffer.appendBuffer() call
--> AppendBufferInternal
--> PrepareAppend
--> OpenIfInEndedState
--> WMSI::UnMmarkEndOfStream
--> CD::UnmarkEndOfStream
DCHECK failure.

2 questions:
1) Why was this next append call allowed to proceed after appendError algorithm had occurred (on some SourceBuffer, maybe a different one on same MediaSource/element)?
2) Why wasn't the state of the ChunkDemuxer 'ended' (previous appendError on *some* WebSourceBufferImpl associated with (I think) that same ChunkDemuxer should have changed it to ended)?