Docker images available at: https://github.com/matrix-org/sliding-sync/pkgs/container/sliding-sync Performance release in preparation for v1.
This version is not compatible with previous versions due to changes in the SQL schema on syncv3_snapshots and syncv3_account_data.
Features:
- to_device messages, OTK counts and device list changes now go through to cilents immediately when they change upstream. Previously they were piggybacked off other data.
Bug fixes:
- When a range is shrinked from both ends, we would incorrectly send back
INVALIDATE twice for the same range (the higher end). This was caused by
improper usage of Gofor .. rangeloops. - When a range is changed to a window completely outside the room list
(e.g 5 rooms,range=[10,15]) we would incorrectly send backSYNC 10,15
with a single room ID, the last room in the list. This was caused by
boundary check code in SliceInto capping ranges to the last element if
the index positions are > room list. However in this case, this causes
the last room to be returned.
Improvements:
- The proxy now logs
content["org.matrix.msgid"]fields if they are present on to-device messages at DEBUG level. - The proxy now warns when clients fail to increase their to-device since token.
- Proxy startup times have been improved with optimised SQL queries.
- The to_device table has had an extra index added to improve performance.
- When >1 poller exists for the same user, the 2nd+ poller wil do their initial sync request filtering out all rooms in an effort to get to_device msgs more rapidly.
- BREAKING: Add stream IDs to account data table.
- BREAKING: Split the event nid array for snapshots into two: membership events and other events. This is used to massively improve performance of looking up
required_statewhen the filter does not include anym.room.memberevents.