Release 2.7.8

Release 2.7.7

Release 2.7.6

Release 2.7.5

Release 2.7.4

Release 2.7.3

[2.7.2] - 2021-10-08

Added

- 'Make all instances writeable' configuration field can be hidden via
  frontend-core's `set_variable` feature or at runtime.

- New `get_issues` callback in role API to collect user-defined issues.
  The issues are gathered from the enabled roles only (present in
  `service-registry`).

- Allow disabling built-in HTTP "admin" user:

  * by specifying `auth_builtin_admin_enabled: false` in the `instances.yml`;
  * using `TARANTOOL_AUTH_BUILTIN_ADMIN_ENABLED=false` environment variable;
  * permanently in `init.lua`:

```lua
-- init.lua

require('cartridge.auth-backend').set_builtin_admin_enabled(false)
cartridge.cfg({
    auth_backend_name = 'cartridge.auth-backend',
    ...
})
```

Changed

- Make built-in HTTP "admin" user a part of default auth backend. Custom
  backends are free of it now.

Fixed

- Eliminate unnecessary transactions after the restart before the replication
  sync. This reduces the chance the hardware restart leads to WAL corruption
  ([#1546](<https://github.com/tarantool/cartridge/issues/1546)).

- Fix `net.box` clients compatibility with future tarantool 2.10 versions.

- Fix vshard rebalancer broken by roles reload.

[2.7.1] - 2021-08-18

Fixed

- Compatibility with Tarantool 2.9 (update ``errors`` dependency to 2.2.1).

[2.7.0] - 2021-08-10

Added

- New suggestion to restart replication. Whenever the replication isn't
  running and the reason isn't in the dead upstream, Cartridge will
  show the corresponding banner in WebUI.

- More server details in WebUI: membership, vshard-router, and
  vshard-storage.

- Roles are stopped with the `on_shutdown` trigger where it's supported
  (in Tarantool 2.8+).

- New `cartridge.cfg` options:

  - `webui_prefix` (default: `""`) allows to modify WebUI routes.
  - `webui_enforce_root_redirect` (default: `true`) manage redirection.

  To sum up, now they look as follows:

  - `<PREFIX>/admin/`;
  - `<PREFIX>/admin/api`;
  - `<PREFIX>/admin/config`;
  - `<PREFIX>/admin/cluster/*`;
  - `<PREFIX>/static/*`;
  - `<PREFIX>/login`;
  - `<PREFIX>/logout`;
  - `/` and `<PREFIX>/` redirect to `/<PREFIX>/admin` (if enabled).

- Support `rebalancer_max_sending` vshard option.

- New `validate_config` method in GraphQL API.

- Add `zone` and `zone_distances` parameters to test helpers.

Changed

- Merge "Schema" and "Code" pages. Also, allow validating all files,
  not only the `schema.yml`.

- Allow expelling a leader. Cartridge will appoint a new leader
  according to the failover priority from the topology.

- Add default `pool.map_call` timeout 10 seconds.

- Forbid starting an instance absent in `instances.yml`.

- Update `errors` dependency to 2.2.0 with a new method
  `errors.netbox_wait_async` to wait for `netbox.future` result.

- Update `membership` dependency to 2.4.0 ([Changelog]
  (https://github.com/tarantool/membership/releases/tag/2.4.0)).

- Update `ddl` dependency to 1.5.0 which supplements the clusterwide
  config with an example schema ([Changelog]
  (https://github.com/tarantool/ddl/releases/tag/1.5.0)).

- Update `vshard` to 0.1.18 ([Changelog]
  (<https://github.com/tarantool/vshard/releases/tag/0.1.18)).

Fixed

- Leaders replaced during stateful failover can be expelled now.

- Make failover logging more verbose.

- Fix hot-reload for roles that leave gaps in httpd routes.

- Check user e-mail uniqueness when editing.

- Expelled instances are removed from the `_cluster` space.

- Fix `get_enabled_roles` to work w/o arguments.

- Don't default to syslog driver unless `/dev/log` or `/var/run/syslog`
  are available.

- Fix inappropriate consistency timeout that led to "Timed out" error
  during forceful leader promotion.

- Support automatic parsing of Tarantool Enterprise box options
  `audit_log` and `audit_nonblock`.

- Instance won't suspect any members during `RecoveringSnapshot` and
  `BootstrappingBox`.

Enhanced in WebUI

- Allow to blacklist subpages for complex modules.

- Fix notifications displaying. Close it by clicking anywhere. Keep it
  open while the mouse is over.

- Various styles enhancements.

[2.6.0] - 2021-04-26

Added

- Update vshard to 0.1.17.
  (`Changelog <https://github.com/tarantool/vshard/releases/tag/0.1.17>`_).

- Update graphql to 0.1.1.
  (`Changelog <https://github.com/tarantool/graphql/releases/tag/0.1.1>`_).

- New test helper: ``cartridge.test-helpers.stateboard``.

- New ``failover`` option in the cluster test helper for easier failover setup.

Changed

- Move DDL related code out of Cartridge and ship it as a permaent role in the
  ddl rock. No observable functionality is affected. The roles remains
  registered implicitly. Nonetheless it's recomended to add it explicitly to
  ``cartridge.cfg({roles = {'cartridge.roles.ddl-manager'}})`` (if it's
  actually used) as this implicity may be removed in future.

Fixed

- Fix unclear timeout errors in case of ``InitError`` and ``BootError`` states.

- Fix inconsistency which could occur while longpolling stateboard in unstable
  networks.

- Increase timeout for the ``validate_config`` stage from 1 to 10 seconds.
  It afftected ``config_patch_clusterwide`` in v2.5, mostly on large clusters.

Enhanced in WebUI

- Highlight if file name exists in file create/rename mode on Code page.

[2.5.1] - 2021-03-24

Added

- Extend GraphQL `issues` API with `aliens` topic. The issues warns if
  two separate clusters share the same cluster cookie.

- Enhance error messages when they're transferred over network. Supply
  it with the connection URI.

Fixed

- Don't skip two-phase commit prematurely. From now on, the decision to
  skip the `apply_config` is made by every instance individually. The
  validation step is never skipped.

- Avoid WebUI and `pool.map_call` requests hanging because of network
  connection problems.

- Fix unclear "Timeout exceeded" error. It affects v2.5.0 two-phase
  commit when an instance is stuck in `ConfiguringRoles` state.

- Make the "Replication isn't running" issue critical instead of a
  warning.

[2.5.0] - 2021-03-05

Issues and suggestions

- Show an issue when `ConfiguringRoles` state gets stuck for more than 5s.

- Show a suggestion to disable broken instances.

- Show a suggestion to force reapply clusterwide configuration.

Configuration options

- New `cartridge.cfg` option `webui_enabled` (default: `true`). Otherwise,
  HTTP server remains operable (and GraphQL too), but serves user-defined
  roles API only.

- New `cartridge.cfg` option `http_host` (default: `0.0.0.0`) which allows
  to specify the bind address of the HTTP server.

Miscellaneous enhancements

- Allow observing cluster from an unconfigured instance WebUI.

- Introduce a new graphql parser (`libgraphqlparser` instead of `lulpeg`).
  It conforms to the newer GraphQL specification and provides better error
  messages. The "null" literal is now supported. But some other GraphQL
  expressions are considered invalid (e.g. empty subselection).

- Properly handle etcd index updates while polling stateful failover updates.
  The problem affected long-running clusters and resulted in flooding logs
  with the "Etcd cluster id mismatch" warnings.

- Refactor two-phase commit (`patch_clusterwide`) logics: don't use hardcoded
  timeout for the `prepare` stage, move `upload` to a separate stage.

- Eliminate GraphQL error "No value provided for non-null ReplicaStatus" when
  a replica is removed from the `box.space._cluster`.

- Allow specifying server zone in `join_server` API.

- Don't make formatting ugly during config upload.

- Hide the bootstrap button when it's not necessary (e.g. before the cluster
  is bootstrapped, and in vshardless cluster too).

- Properly display an error if changing server zone fails.

- Fix disabled instances style.

Release 2.4.0

Fencing and failover

- The new fencing feature protects a replicaset from the presence of
  multiple leaders when the network is partitioned. It forces an
  isolated leader to become read-only.

- New failover parameter `failover_timout` specifies the time (in
  seconds) used by membership to mark `suspect` members as `dead` which
  triggers failover. The default value of `failover_timeout` increased
  from 3 to 20 seconds (**important change**).

- The `rpc_*` functions now consider `suspect` members as healthy to be
  in agreement with failover (**important change**).

- All these parameters, `fencing_enabled`, `fencing_pause`,
  `fencing_timeout`, are available for customization via Lua and
  GraphQL API, and in WebUI too.

Zones and zone distances

- Add support of replica weights and zones via a cluster-wide config new
  section `zone_distances` and a server parameter `zone`.

- Update server details dialog in WebUI.

Issues and suggestions

- Fix `upstream.idle` issue tolerance to avoid unnecessary warnings
  "Replication: long idle (1 > 1)".

- Indicate config checksum mismatch in the issues list.

- Indicate the change of `arvertise_uri` in the issues list.

- Show an issue if the cluster-wide config is locked on an instance.

- Show a suggestion to heal the cluster after relocating servers and
  changing their `advertise_uri`.

- Don't get stuck in `ConnectingFullmesh` state when an instance is
  restarted with a different `advertise_uri`. Also, keep the "Server
  details" dialog in WebUI operable in this case.

- Allow applying config when an instance is in `OperationError` state.
  It doesn't cause a loss of quorum anymore.

Hot-reload

- Add new feature for hot reloading roles code without restarting an
  instance: `cartridge.reload_roles()`. The feature is experimental
  and should be enabled explicitly:
  `cartridge.cfg({roles_reload_allowed = true})`. Use it with caution.

Miscellaneous enhancements

- New `cartridge.cfg` option `swim_broadcast` to manage instances
  auto-discovery on start. Default: true.

- New argparse options for tarantool 2.5+: `replication_synchro_quorum`,
  `replication_synchro_timeout`, `memtx_use_mvcc_engine`.

- Stop vshard fibers when the corresponding role is disabled.

- Make `console.listen` error more clear when `console_sock` exceeds
  `UNIX_PATH_MAX` limit.

- Allow removing spaces from DDL schema for the sake of `drop`
  migrations.

- Make DDL schema validation stricter. Forbid redundant keys in schema
  top-level and make `spaces` mandatory.

- Refresh interval and stat refresh period variables can be customized
  via frontend-core `set_variable` feature or at runtime.

- Properly display errors on WebUI pages "Users" and "Code".

Release 2.3.0

Added

- When failover mode is stateful, all manual leader promotions will be
  consistent: every instance before becoming writable performs
  `wait_lsn` operation to sync with previous one. If consistency
  couldn't be reached due to replication failure, a user could either
  revert it (promote previous leader), or force promotion to be
  inconsistent.

- Early logger initialization (for Tarantool > 2.5.0-100, which
  supports it).

- Add `probe_uri_timeout` argparse option responsible for retrying
  "Can't ping myself" error on startup.

- New test helper: `cartridge.test-helpers.etcd`.

- Support `on_push` and `on_push_ctx` options for `cartridge.rpc_call()`.

- Changing users password invalidates HTTP cookie.

- Support GraphQL default variables
  (https://graphql.org/learn/queries/#default-variables).

Fixed

- Eventual failover may miss an event while roles are being reconfigured.

- Compatibility with pipe logging, see
  https://github.com/tarantool/tarantool/issues/5220.

- Non-informative assertion when instance is bootstrapped with a
  distinct `advertise_uri`.

- Indexing `nil` value in `get_topology()` query.

- Initialization race of vshard storage which results in
  `OperationError`.

- Lack of vshard router reconnection attempts to the replicas.

- Make GraphQL syntax errors more clear.

- Better `errors.pcall()` performance, `errors` rock updated to v2.1.4.

Enhanced is WebUI

- Show instance names in issues list.

- Show app name in window title.

- Add "Force leader promotion" button in stateful failover mode.

- Indicate consistent switchover problems with a yellow leader flag.

Release 2.2.1

Fixed

- Eventual failover may miss an event while roles are being
  reconfigured.

Release 2.2.0

Added

- When running under systemd use `<APP_NAME>.<INSTANCE_NAME>` as
  default syslog identity.

- Support `etcd` as state provider for stateful failover.

Changed

- Improve rocks detection for feedback daemon. Besides cartridge version
  it now parses manifest file from the `.rocks/` directory and collects
  rocks versions.

- Make `uuid` parameters optional for test helpers. Make `servers`
  option accept number of servers in replicaset.

Enhanced in WebUI

- Prettier errors displaying.

- Enhance replicaset filtering by role / status.

Release 2.1.2

Fixed

- Avoid trimming `console_sock` if it's name is too long.

- Fix file descriptors leak during box recovery.

- Support `console_sock` option in stateboard as well as notify socket
  and other box options similar to regular cartridge instances.

Release 2.1.1

Fixed

- Fix route mapping in WebUI

Release 2.1.0

Added

- Implement stateful failover mode. You can read more in "Failover
  architecture" documentation topic (see `failover.md`).

- Respect `box.cfg` options `wal_dir`, `memtx_dir`, `vinyl_dir`. They
  can be either absolute or relative - in the later case it's calculated
  relative to `cartridge.workdir`.

- New option in `cartridge.cfg({upgrade_schema=...})`
  to automatically upgrade schema to modern tarantool version
  (only for leader). It also has been added for `argparse`.

- Extend GraphQL `issues` API with various topics: `replication`,
  `failover`, `memory`, `clock`. Make thresholds configurable via
  argparse.

Changed

- Make GraphQL validation stricter: scalar values can't have
  sub-selections; composite types must have sub-selections; omitting
  non-nullable arguments in variable list is forbidden. Your code **may
  be affected** if it doesn't conform GraphQL specification.

- GraphQL query `auth_params` returns "fullname" (if it was specified)
  instead of "username".

- Update `errors` dependency to 2.1.3.

- Update `ddl` dependency to 1.1.0.

Deprecated

Lua API:

- `cartridge.admin_get_failover` -> `cartridge.failover_get_params`
- `cartridge.admin_enable/disable_failover` -> `cartridge.failover_set_params`

GraphQL API:

- `query {cluster {failover} }` -> `query {cluster {failover_params {...} } }`
- `mutation {cluster {failover()} }` -> `mutation {cluster {failover_params() {...} } }`

Fixed

- Properly handle nested input object in GraphQL:
```graphql
  mutation($uuid: String!) {
    cluster { edit_topology(servers: [{uuid: $uuid ...}]) {} }
  }
```

- Show WebUI notification on successful config upload.

- Repair GraphQL queries `add_user`, `issues` on uninitialized instance.

Enhanced in WebUI

- Show "You are here" marker.

- Show application and instance names in app title.

- Indicate replication and failover issues.

- Fix bug with multiple menu items selected.

- Refactor pages filtering, forbid opening blacklisted pages.

- Enable JS chunks caching.

Release 2.1.0

Added

- Implement stateful failover mode. You can read more in "Failover
  architecture" documentation topic (see `failover.md`).

- Respect `box.cfg` options `wal_dir`, `memtx_dir`, `vinyl_dir`. They
  can be either absolute or relative - in the later case it's calculated
  relative to `cartridge.workdir`.

- New option in `cartridge.cfg({upgrade_schema=...})`
  to automatically upgrade schema to modern tarantool version
  (only for leader). It also has been added for `argparse`.

- Extend GraphQL `issues` API with various topics: `replication`,
  `failover`, `memory`, `clock`. Make thresholds configurable via
  argparse.

Changed

- Make GraphQL validation stricter: scalar values can't have
  sub-selections; composite types must have sub-selections; omitting
  non-nullable arguments in variable list is forbidden. Your code **may
  be affected** if it doesn't conform GraphQL specification.

- GraphQL query `auth_params` returns "fullname" (if it was specified)
  instead of "username".

- Update `errors` dependency to 2.1.3.

- Update `ddl` dependency to 1.1.0.

Deprecated

Lua API:

- `cartridge.admin_get_failover` -> `cartridge.failover_get_params`
- `cartridge.admin_enable/disable_failover` -> `cartridge.failover_set_params`

GraphQL API:

- `query {cluster {failover} }` -> `query {cluster {failover_params {...} } }`
- `mutation {cluster {failover()} }` -> `mutation {cluster {failover_params() {...} } }`

Fixed

- Properly handle nested input object in GraphQL:
```graphql
  mutation($uuid: String!) {
    cluster { edit_topology(servers: [{uuid: $uuid ...}]) {} }
  }
```

- Show WebUI notification on successful config upload.

- Repair GraphQL queries `add_user`, `issues` on uninitialized instance.

Enhanced in WebUI

- Show "You are here" marker.

- Show application and instance names in app title.

- Indicate replication and failover issues.

- Fix bug with multiple menu items selected.

- Refactor pages filtering, forbid opening blacklisted pages.

- Enable JS chunks caching.

Release 2.0.2

**Added**

- Expose membership options in `argparse` module (edit them with
  environment variables and command-line arguments).
- New internal module to handle `.tar` files.

Lua API:

- `cartridge.cfg({webui_blacklist = {'/admin/code', ...}})`: blacklist
  certain WebUI pages.
- `cartridge.get_schema()` referencing older `_G.cartridge_get_schema`.
- `cartridge.set_schema()` referencing older `_G.cartridge_set_schema`.

GraphQL API:

- Make use of GraphQL error extensions: provide additional information
  about `class_name` and `stack` of original error.
- `cluster{ issues{ level message ... }}`: obtain more details on
  replication status
- `cluster{ self {...} }`: new fields `app_name`, `instance_name`.
- `servers{ boxinfo { cartridge {...} }}`: new fields `version`,
  `state`, `error`.

Test helpers:

- Allow specifying `all_rw` replicaset flag in luatest helpers.
- Add `cluster({env = ...})` option for specifying clusterwide
  environment variables.

**Changed**

- Remove redundant topology availability checks from two-phase commit.
- Prevent instance state transition from `ConnectingFullmesh` to
  `OperationError` if replication fails to connect or to sync. Since now
  such fails result in staying in `ConnectingFullmesh` state until it
  succeeds.
- Specifying `pool.connect()` options `user`, `password`,
  `reconnect_after` are deprecated and ignored, they never worked as
  intended and will never do. Option `connect_timeout` is deprecated,
  but for backward compatibility treated as `wait_connected`.

**Deprecated**

Lua API:

- `cartridge.admin_get_failover` -> `cartridge.failover_get_params`
- `cartridge.admin_enable/disable_failover` ->
  `cartridge.failover_set_params`

GraphQL API:

- `query {cluster {failover} }` ->
  `query {cluster {failover_params {...} } }`
- `mutation {cluster {failover()} }` ->
  `mutation {cluster {failover_params() {...} } }`

**Fixed**

- Fix DDL failure if `spaces` field is `null` in input schema.
- Check content of `cluster_cookie` for absence of special
  characters so it doesn't break the authorization.
  Allowed symbols are `[a-zA-Z0-9_.~-]`.
- Drop remote-control connections after full-featured `box.cfg` becomes
  available to prevent clients from using limited functionality for too
  long. During instance recovery remote-control won't accept any
  connections: clients wait for box.cfg to finish recovery.
- Update errors rock dependency to 2.1.2: eliminate duplicate stack
  trace from `error.str` field.
- Apply `custom_proc_title` setting without waiting for `box.cfg`.
- Make GraphQL compatible with `req:read_cached()` call in httpd hooks.
- Avoid "attempt to index nil value" error when using rpc on an
  uninitialized instance.

**Enhanced in WebUI**

- Add an ability to hide certain WebUI pages.
- Validate YAML in code editor WebUI.
- Fix showing errors in Code editor page.
- Remember last open file in Code editor page.
  Open first file when local storage is empty.
- Expand file tree in Code editor page by default.
- Show Cartridge version in server info dialog.
- Server alias is clickable in replicaset list.
- Show networking errors in splash panel instead of notifications.
- Accept float values for vshard-storage weight.