wireapp/wire-server v2024-02-12

2024-02-12 (Chart Release 4.40.0)

on GitHub

Release notes

• The settings setDisabledAPIVersions (brig) and disabledAPIVersions (in cannon, cargohold, galley, gundeck, proxy, and spar) are now required.
  The default defined in charts/<service>/values.yaml is set to [ development ] and disables all development API versions.
  For more information see https://docs.wire.com/developer/reference/config-options.html#disabling-api-versions (#3772)

• The mls team feature now has a lock status. If the current installation has default settings defined in wire-server's values.yaml, the lockStatus has to be added, e.g.:

  mls:
    defaults:
      status: enabled
      config:
        protocolToggleUsers: []
        defaultProtocol: mls
        allowedCipherSuites: [1]
        defaultCipherSuite: 1
        supportedProtocols: [proteus, mls] # must contain defaultProtocol
      lockStatus: locked

(#3681)

• • Replace fake-sqs with ElasticMQ

  ElasticMQ is an actively maintained project, fake-sqs hasn't seen a commit since
  2018. This is not expected to have any noticeable effect on deployments that
  don't have any extra configurations for the SQS queues. If the fake-aws-sqs
  chart had configured custom queue names, they have couple of extra limitations:

  • The queue names must only contain alphanumeric characters and hyphens.
  • The FIFO queue names must end in .fifo. (#3750)

• Upgrade team-settings version to 4.15.1-v0.31.19-0-ee1dbce (#2180)

• Upgrade webapp to 2023-12-11-production.0-v0.31.17-0-1e91445

  Beside using up-to-date versions in Helm charts is generally beneficial,
  this version also provides multi-ingress support. (#3803)

• Upgrade webapp version to 2024-01-22-production.1-v0.31.17-0-7f83dbe (#2302)

API changes

• Un-verified users can no longer upload assets (#3604)

• • Remove IP request header from add-client endpoint.
  • Remove longitude and latitude response fields from add-client, list-clients, and get-client endpoints.

  These are not considered breaking changes, since clients are not using this information. (#3792)

• Create new API version V6 and finalize V5 (#3815)

• Block changes of userDisplayName, userHandle in mlsE2EI-enabled teams on the backend without SCIM; report "managed_by" == "scim" in GET /self, but only there (#3827)

• The federation API can now be versioned. Multiple versions of an RPC can be defined on the same path. After version negotiation, the federation client now sets the X-Wire-API-Version header, and federator propagates it to the destination service. (#3762)

• Improved formatting of federation errors. No extra copy of the response body, and nested errors are now part of the JSON structure, not quoted inside the message. (#3742)

• New endpoint for replacing MLS key packages in bulk: PUT /mls/key-packages/self/:client. It replaces all existing key packages that match the given ciphersuites with the new key packages provided in the body. (#3654)

Features

• The lifetime of conversation guest links is now configurable (#3796)

• Events for a member update, join and leave are not sent to everyone in the team any longer. Only team admins get them. (#3703)

• Allowlist for who on cloud can connect to on-prem:

  • Internal API to configure allowlist
  • Restrict federated user search according to team federation policy
    (#3697, #3732, #3758)

• The mls team feature now has a lock status (#3681)

• add a uniform timeout to the integration test-suite set by the environment variable TEST_TIMEOUT_SECONDS with a default of 10 seconds if the variable isn't set (#3692)

• Apply team-level federation policies when establishing and updating user connections (#3774)

• Introduce a feature flag that controls whether the limited event fanout should be used when a team member is deleted (#3797)

• Send a conversation.member-leave event to team admins for each conversation the deleted team member used to be part of (#3790)

• Allow the configuration of TLS-secured connections to Cassandra. TLS is used
  when a certificate is provided. This is either done with
  --tls-ca-certificate-file for cli commands or the configuration attribute
  cassandra.tlsCa for services. In Helm charts, the certificate is provided as
  literal PEM string; either as attribute cassandra.tlsCa (analog to service
  configuration) or by a reference to a secret (cassandra.tlsCaSecretRef.) (#3587)

• [SFT chart] Add option to enable serviceMonitor to scrape prometheus metrics (#3770)

Bug fixes and other updates

• galley's DB migrations fixed (#3680)

• The X509 client identity parser supports a new format: wireapp://{userid}!{deviceid}@{host} (#3808)

• Updated demo-smtp Helm chart from deprecated docker image namshi/smtp to ixdotai/smtp (#3791)

• External partners search restriction enforced by backend (#3708)

• File upload size is now limited to 100 MiB (#3752)

• Fix a bug where non-team conversation members that are remote would not get a conversation.member-leave event (#3745, #3764)

• Enforce external partner permissions on the backend (#3788)

• Various improvements around LH policy conflict detection:

  • Fix LH policy conflict detection logic when posting messages
  • Better policy conflict error messages (distinguish between old clients and missing consent)
  • Add first LH scaffolding and tests to /integration
  • Annotate some API functions in /integration with links to openapi3 docs (#3773)

• Do not match on the Accept header for service provider endpoints with no response body (#3766)

• Guests should not be added to conversations that are under legalhold (#3853)

• Intra-service calls from brig to galley's public API are now aware of disabled API versions (#3863)

• fix Helm pretty-printer for disabledAPIVersions (#3877)

• Adjust the requested memory and upper bound limit of nginz pods in the related Helm chart. (We experienced OOM errors with the old settings.) (#3821)

• don't use shell when communicating with mls-test-cli, move flaking brig tests over to new integration testsuite (#3701)

• set notificationTimeOut to 28 days, make it legible (#3714)

• Update coturn image with bugfix to its prestop-hook from wireapp/coturn#10 to allow coturn pods to terminate once their traffic has drained. (#3872)

• Extra remove proposals were being sent when a user was removed from a conversation (#3672)

• Remove client check for subconversations (#3677)

• Ensure that SCIM can find users even after the team admin has changed the SAML issuer for the user. (#3747)

• addClient used the internal brig API in the integration testsuite when it should use the public one (#3869)

• Ensure that HTTP 1.1 connections are grafully closed

  To fix this warp had to be patched to fix the bug upstream: yesodweb/wai#958 (#3775)

Documentation

• Fix missing code sections on docs.wire.com, notably on "configuring TLS" page. (#3839)

• Swagger generation no longer adds tags containing information about federation calls.

  Added the federation calling graph to the Federation API Conventions page. (#3674)

• Backend-to-backend OpenApi Docs added (#3666)

• Documentation for creating a new API version updated (#3817)

• Update documentation of MLS group ID (#3705)

• Turn long summaries in openapi documentation into descriptions (#3706)

• update the build instructions for wire-server (#3854)

Internal changes

• stern/backoffice PUT /teams/{teamId}/features/conferenceCalling fixed (#3723)

• Removed client ID conversion round trip (#3727)

• Migrate to Servant the Galley conversation internal endpoints (#3718)

• The development API version is now disabled by default (#3772)

• Attempt to fix flaky integration test provider.service.delete (#3689)

• The fedcalls tool no longer walks the Swagger/OpenAPI structure when generating call graphs. These graphs are now generated directly from the Servant API types. (#3674, #3691)

• Increased ingress payload size from 256k to 512k (#3756)

• Request tracing across federated requests (#3765)

• upgrade nixpkgs to upgrade haskell-language-server (#3650)

• upgrade the GHC version to GHC 9.4 (#3679)

• Removed APNS_VOIP code. (APNS_VOIP is a native push notification channel which we aren't using anymore.) (#3695)

• Improve error logs (#3782)

• Migrating tests for Cargohold to the new integration test suite. (#3741)

• Fix calendar integration setting in backoffice / stern (#3761)

• Reply-Nonce is added to Access-Control-Expose-Headers (#3729)

• Add custom feature flag; only supported for some on-prem installations; locked & disabled by default (#3779)

• Improved how tests are automatically extracted from the integration test suite.

  The test extractor parser has been improved to handle block comments, and to more strictly check for Haddock documentation for each test. (#3749)

• Additional logging on user/team suspension (#3795)

• cleanup the haskell-pins

  • remove many pins
  • remove many overrides
  • restructure the files such that it's easier to see what is going on (#3814)

• Version of rusty-jwt-tools bumped to v0.8.0 (#3805)

• Feature enforceFileDownloadLocation lockstatus can be set with basic auth on staging (#3802)

• Version of rusty-jwt-tools bumped to v0.8.5 (#3820)

• Translate integration tests: manually add / delete LH device (#3830)

• adds a new executable, hs-run, to quickly run haskell scripts (#3716)

• Represent client IDs as Word64 internally (#3713)

• Allow to install the coturn chart multiple times in multiple namespaces on the same cluster. (#3698)

• For some rust packages (cryptobox and libzauth-c), we now use crate2nix as a build tool, rather than the more coarse and FOD-based nixpkgs rustPlatform.buildRustPackage approach. (#3686)

• Delete shell.nix. It has been broken for quite some time. The supported way to get a development nix environment is to use direnv. (#3726)

• Deploy a backend with federation API V0 while setting up services for local testing (#3719)

• Improve integration test coverage (#3757)

• Increase timeout for waiting for SQS notifications in galley's integration tests (#3699)

• Simplify process spawning of dynamic backends in integration tests (#3759)

• More robust consuming of MLS messages: the behaviour of sendAndConsumeMessage and sendAndConsumeCommitBundle is changed to actually wait for those messages on the client's websocket (#3671)

• Update group state after application message (#3678)

• bump the nixpkgs version to allow updating curl (#3781)

• Simplify the definition of the servant notification API (#3685)

• Start refactoring code into subsystems, first subsystem being the NotificationSubsystem. (#3786)

• Remove apply-refact from CI image

  This gets rid of GHC in the image, making the image smaller. (#3712)

• Refactor getOptions (#3707)

• Restored Brig memory quota to 512mb down from 1gb. (prev bump #3751) (#3806)

• Add tool to analyse test results in junit/ant xml format (#3652)

• updated annotation for enabling Topology Aware Routing to service.kubernetes.io/topology-mode for k8s 1.27+ (#3878)

• replace runAsNonRoot with runAsUser and runAsGroup 1000 (#3826)

• Update SFTD default to 4.0.10 and its nginx to 1.25.3. (#3768)

• add a Makefile target to make it possible to upload a bom of all services to s3 on every CI run (#3744)

• Upload bill-of-material (BOM) files directly to the Dependency Tracker via REST.
  This eases the life of the security team and prevents cluttering our release
  artifact page. (#3810)

• Passively migrate user passwords from scrypt to argon2id.

  By passively we mean that whenever a user re-enters their passwords, if it was hashed using scrypt, it is then rehashed using argon2id and stored as such.
  If that user has a legacy short password (under 8 characters in length), it does not migrate to argon2id. (#3720)

Federation changes

• Define a few tests for adding members to an MLS conversation when unreachable backends are involved (#3673)

• Make sure that remote users can be added to both a Proteus and an MLS conversation when other users are unreachable (#3688)