I've been looking through some of the JNI errors we see in issue 726426.

The "global reference table overflow" error appears to be due to leaking remotemediaplayerbridge. 

Here's the set of crash reports:
https://crash.corp.google.com/browse?q=product.name%3D%27Chrome_Android%27%20AND%20custom_data.ChromeCrashProto.ptype%3D%27browser%27%20AND%20product.Version%3D%2760.0.3112.107%27%20OMIT%20RECORD%20IF%20SUM(special_protos.android_info.logcat_errors%20LIKE%20%27%25global%20reference%20table%20overflow%25%27)%20%3D%200&sql_dialect=dremelsql&ignore_case=false&enable_rewrite=false&omit_field_name=&omit_field_value=&omit_field_opt=&stbtiq=&reportid=99fb231f24255306&index=5#7

I've seen a few cases with thousands of Surface object allocations and in this particular case 1949 RemoteMediaPlayerBridge instances which seems wrong.

I suspect we're not freeing in some error cases (based on some of the errors I'm seeing in the logs)

Mounir: can you triage?