This repository has been archived by the owner on Apr 26, 2024. It is now read-only.
Increase MAX_EVENTS_BEHIND for replication clients #6967
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
erikjohnston
approved these changes
Feb 21, 2020
richvdh
added a commit
that referenced
this pull request
Mar 23, 2020
Synapse 1.12.0 (2020-03-23) =========================== No significant changes since 1.12.0rc1. Debian packages and Docker images are rebuilt using the latest versions of dependency libraries, including Twisted 20.3.0. **Please see security advisory below**. Security advisory ----------------- Synapse may be vulnerable to request-smuggling attacks when it is used with a reverse-proxy. The vulnerabilties are fixed in Twisted 20.3.0, and are described in [CVE-2020-10108](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2020-10108) and [CVE-2020-10109](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2020-10109). For a good introduction to this class of request-smuggling attacks, see https://portswigger.net/research/http-desync-attacks-request-smuggling-reborn. We are not aware of these vulnerabilities being exploited in the wild, and do not believe that they are exploitable with current versions of any reverse proxies. Nevertheless, we recommend that all Synapse administrators ensure that they have the latest versions of the Twisted library to ensure that their installation remains secure. * Administrators using the [`matrix.org` Docker image](https://hub.docker.com/r/matrixdotorg/synapse/) or the [Debian/Ubuntu packages from `matrix.org`](https://github.com/matrix-org/synapse/blob/master/INSTALL.md#matrixorg-packages) should ensure that they have version 1.12.0 installed: these images include Twisted 20.3.0. * Administrators who have [installed Synapse from source](https://github.com/matrix-org/synapse/blob/master/INSTALL.md#installing-from-source) should upgrade Twisted within their virtualenv by running: ```sh <path_to_virtualenv>/bin/pip install 'Twisted>=20.3.0' ``` * Administrators who have installed Synapse from distribution packages should consult the information from their distributions. The `matrix.org` Synapse instance was not vulnerable to these vulnerabilities. Advance notice of change to the default `git` branch for Synapse ---------------------------------------------------------------- Currently, the default `git` branch for Synapse is `master`, which tracks the latest release. After the release of Synapse 1.13.0, we intend to change this default to `develop`, which is the development tip. This is more consistent with common practice and modern `git` usage. Although we try to keep `develop` in a stable state, there may be occasions where regressions creep in. Developers and distributors who have scripts which run builds using the default branch of `Synapse` should therefore consider pinning their scripts to `master`. Synapse 1.12.0rc1 (2020-03-19) ============================== Features -------- - Changes related to room alias management ([MSC2432](matrix-org/matrix-spec-proposals#2432)): - Publishing/removing a room from the room directory now requires the user to have a power level capable of modifying the canonical alias, instead of the room aliases. ([\#6965](#6965)) - Validate the `alt_aliases` property of canonical alias events. ([\#6971](#6971)) - Users with a power level sufficient to modify the canonical alias of a room can now delete room aliases. ([\#6986](#6986)) - Implement updated authorization rules and redaction rules for aliases events, from [MSC2261](matrix-org/matrix-spec-proposals#2261) and [MSC2432](matrix-org/matrix-spec-proposals#2432). ([\#7037](#7037)) - Stop sending m.room.aliases events during room creation and upgrade. ([\#6941](#6941)) - Synapse no longer uses room alias events to calculate room names for push notifications. ([\#6966](#6966)) - The room list endpoint no longer returns a list of aliases. ([\#6970](#6970)) - Remove special handling of aliases events from [MSC2260](matrix-org/matrix-spec-proposals#2260) added in v1.10.0rc1. ([\#7034](#7034)) - Expose the `synctl`, `hash_password` and `generate_config` commands in the snapcraft package. Contributed by @devec0. ([\#6315](#6315)) - Check that server_name is correctly set before running database updates. ([\#6982](#6982)) - Break down monthly active users by `appservice_id` and emit via Prometheus. ([\#7030](#7030)) - Render a configurable and comprehensible error page if something goes wrong during the SAML2 authentication process. ([\#7058](#7058), [\#7067](#7067)) - Add an optional parameter to control whether other sessions are logged out when a user's password is modified. ([\#7085](#7085)) - Add prometheus metrics for the number of active pushers. ([\#7103](#7103), [\#7106](#7106)) - Improve performance when making HTTPS requests to sygnal, sydent, etc, by sharing the SSL context object between connections. ([\#7094](#7094)) Bugfixes -------- - When a user's profile is updated via the admin API, also generate a displayname/avatar update for that user in each room. ([\#6572](#6572)) - Fix a couple of bugs in email configuration handling. ([\#6962](#6962)) - Fix an issue affecting worker-based deployments where replication would stop working, necessitating a full restart, after joining a large room. ([\#6967](#6967)) - Fix `duplicate key` error which was logged when rejoining a room over federation. ([\#6968](#6968)) - Prevent user from setting 'deactivated' to anything other than a bool on the v2 PUT /users Admin API. ([\#6990](#6990)) - Fix py35-old CI by using native tox package. ([\#7018](#7018)) - Fix a bug causing `org.matrix.dummy_event` to be included in responses from `/sync`. ([\#7035](#7035)) - Fix a bug that renders UTF-8 text files incorrectly when loaded from media. Contributed by @TheStranjer. ([\#7044](#7044)) - Fix a bug that would cause Synapse to respond with an error about event visibility if a client tried to request the state of a room at a given token. ([\#7066](#7066)) - Repair a data-corruption issue which was introduced in Synapse 1.10, and fixed in Synapse 1.11, and which could cause `/sync` to return with 404 errors about missing events and unknown rooms. ([\#7070](#7070)) - Fix a bug causing account validity renewal emails to be sent even if the feature is turned off in some cases. ([\#7074](#7074)) Improved Documentation ---------------------- - Updated CentOS8 install instructions. Contributed by Richard Kellner. ([\#6925](#6925)) - Fix `POSTGRES_INITDB_ARGS` in the `contrib/docker/docker-compose.yml` example docker-compose configuration. ([\#6984](#6984)) - Change date in [INSTALL.md](./INSTALL.md#tls-certificates) for last date of getting TLS certificates to November 2019. ([\#7015](#7015)) - Document that the fallback auth endpoints must be routed to the same worker node as the register endpoints. ([\#7048](#7048)) Deprecations and Removals ------------------------- - Remove the unused query_auth federation endpoint per [MSC2451](matrix-org/matrix-spec-proposals#2451). ([\#7026](#7026)) Internal Changes ---------------- - Add type hints to `logging/context.py`. ([\#6309](#6309)) - Add some clarifications to `README.md` in the database schema directory. ([\#6615](#6615)) - Refactoring work in preparation for changing the event redaction algorithm. ([\#6874](#6874), [\#6875](#6875), [\#6983](#6983), [\#7003](#7003)) - Improve performance of v2 state resolution for large rooms. ([\#6952](#6952), [\#7095](#7095)) - Reduce time spent doing GC, by freezing objects on startup. ([\#6953](#6953)) - Minor perfermance fixes to `get_auth_chain_ids`. ([\#6954](#6954)) - Don't record remote cross-signing keys in the `devices` table. ([\#6956](#6956)) - Use flake8-comprehensions to enforce good hygiene of list/set/dict comprehensions. ([\#6957](#6957)) - Merge worker apps together. ([\#6964](#6964), [\#7002](#7002), [\#7055](#7055), [\#7104](#7104)) - Remove redundant `store_room` call from `FederationHandler._process_received_pdu`. ([\#6979](#6979)) - Update warning for incorrect database collation/ctype to include link to documentation. ([\#6985](#6985)) - Add some type annotations to the database storage classes. ([\#6987](#6987)) - Port `synapse.handlers.presence` to async/await. ([\#6991](#6991), [\#7019](#7019)) - Add some type annotations to the federation base & client classes. ([\#6995](#6995)) - Port `synapse.rest.keys` to async/await. ([\#7020](#7020)) - Add a type check to `is_verified` when processing room keys. ([\#7045](#7045)) - Add type annotations and comments to the auth handler. ([\#7063](#7063))
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is intended to fix, or at least work around, #2738.
The problem is essentially that when we join a room with more than 10000 state events (ie, a room that has ever had more than 10000 different mxids in it), this creates an entry in the replication stream which consists of 10000 rows. The master then thinks that the worker has got behind and drops the connection. The worker reconnects, but immediately the same problem recurs.
There's some debate about whether it's useful to have a cap at all: it's not like booting the worker off actually fixes anything, so it's more a red flag to indicate that there's a problem (and helps stop the master ending up with massive queues of replication data when the workers get behind).
Still, it's likely that we'll be reworking things in this department soon as part of #6677, so for now let's just make the limit huge and move on.