Skip to content

Latest commit

聽

History

History
509 lines (420 loc) 路 25.8 KB

CHANGELOG.md

File metadata and controls

509 lines (420 loc) 路 25.8 KB
Raw

Release v2.2 (2019-08-23)

Major Topics

Supervisor plugin A new supervisor plugin was introduced, effectively capable to replace supervisord-based environment. The plugin defines a configuration file with a list of programs started by the so-called supervisor agent, or agent-init. More information is available is Ligato documentation for supervisor plugin.

New Features

  • Supervisor-plugin
    • A new management plugin can start other processes, including other agents or VPP. Every process must specify a unique name, executable-path and executable-args (arguments) if they are used.
    • Base logging is done to the standard output/err. Instead, all the output can be written to the file, defining file path in logfile-path within the supervisor config file.
    • Every process can now have the CPU affinity set. Use cpu-affinity-mask field to set hexadecimal CPU mask (the same format as Linux command taskset). Another field, cpu-affinity-setup-delay can postpone CPU affinity configuration by a given time.
    • Hooks are executed for specific process events, like a change of the process state. Every hook is executed in supervisor environment with several environment variables set. See the documentation for more information.

Improvements

  • ETCD
    • Now supports atomic operations with interface BytesBrokerWithAtomic. Also the new API method CompareAndSwap was added.
    • Added option to use 'expand' environment variable which replaces $var in received JSON data.
    • Option to set own serialized as a plugin dependency.
    • New method RawAccess() accesses data in the database as bytes (i.e. not formatted as protobuf).
  • Process-manager
    • Now uses a buffered writer to put logs to either stdout/stderr or to file. The file is written if the field logfile-path is set.

Release v2.1 (2019-05-09)

Improvements

  • Datasync
    • Added new option WithClientLifetimeTTL. This option defines put operation for the lifetime of a client (the TTL is not renewed if the client is closed).
  • ETCD
    • Support for election mechanism, which determines leader instance for given prefix. Multiple ETCD instances can now compete for leadership.
  • Bolt
    • Some improvements to Bolt watchers were done in order to reduce a number of active go routines

Fixed Bugs

  • BoltDB watcher: keys sent to the close channel are now properly prefixed and delivered to the correct go routine so they will not be ignored

Documentation

Release v2.0 (2019-04-02)

Breaking Changes

  • The ChangeEvent interface was modified. Before, it returned single value of type ProtoWatchResp, now it contains a new method GetChanges() which return a list of values of that type.

New Features

  • Datasync
    • The ChangeEvent interface provides a new method GetContext returning a context associated with the given event.
    • The ResyncEvent interface also provides a new method called GetContext returning a context associated with the given resync event.
    • Resync time duration is shown in milliseconds
  • IdxMap
    • New method ListFields(string) in the NamedMapping interface providing a map of fields associated with the item identified by the named parameter known as secondary indexes
  • GRPC
    • Added authentication support for the GRPC plugin
  • Probe
    • Support for non-fatal errors. The probe plugin keeps a lst of non-fatal plugins. Errors reported from the non-fatal plugin are effectively ignored in the vpp-agent overall status.
  • BoltDB
    • The config file now has an option to filter duplicated notifications.

Improvements

  • Agent
    • DumpStackTraceOnTimeout is now disabled by default with possibility to enable it via the environment variable DUMP_STACK_ON_TIMEOUT.
    • The agent logs the last plugin in case it fails to start because of the timeout
    • Signal received during startup closes the agent instance
  • FileDB
    • If the child process is not detached, the PDeathSignal is set preventing the child process to hang
    • The process watcher is started on process start, rather than on process creation
    • A new or an attached process can be created with custom I/O writer
    • Support for environment variables for processes
  • Resync
    • The resync timeout was split into two values, for ACK timeout set to 10 second (up from 5) and for ACCEPT timeout set to 1 second.

Fixed Bugs

  • The ETCD plugin watch is now properly closed using a context
  • The plugin lookup works correctly with interface types where the inner type is slice or array of plugins
  • PropagateResync now waits at the result of the resync event instead of returning immediately
  • Used jsonpb to fix protobuf marshalling/unmarshalling
  • Plugin config directory paths with ".." are now handled properly.
  • If channel used by process manager is closed by the used (which should not be done), the plugin recovers

Documentation

  • Added "beginner" tutorials (hello world, plugin dependencies, REST handler, KV Store, plugin lookup) with examples.

Release v1.7 (2018-12-12)

Major Topics

FileDB

With the new release, the cn-infra introduces a new feature where the filesystem itself can be used as a key-value data store. Defined files or directories can be used to store configuration items the same way as any other database (ETCD, Redis, etc.). The data have to have prescribed format but otherwise, they respect proto models from particular plugins. Configuration files can be of JSON or YAML type. The fileDB plugin supports watcher, configuration status update or resync - features known from other key-value data stores.

Process management plugin

A new plugin was added to extend cn-infra functionality - a process manager. The plugin can be used to create, start and monitor processes. Begin with creating new process instance like NewProcess(<cmd>, <options>...) and start it with Start(). The process manager supports all the common commands like restart, termination, kill, wait for process completion, check process liveness or watch its status. More advanced features or options allow to detach the process from a parent, preserve it when an application is restarted or automatic cleanup of zombie processes.

New features

  • Added new plugin allowing usage of a filesystem as the key-value data store. More details in the readme
  • Added new plugin for external process management. More information in the readme
  • StatusCheck plugin has a new option to define non-fatal plugins. A registered plugin marked as non-fatal is not propagated into overall status.
  • Added watcher for BoltDB database. The BoltDB now supports all the watcher options as other key-value data store plugins.

Examples

  • Redis-lib simple example was updated and changed, now clearly demonstrating Redis plugin options and usage. The Redis "airport" was rewritten, the functionality remained unchanged, but it should be a lot easier to understand the code.

Other

  • Agent start/stop error now prints stack trace

Release v1.6 (2018-10-04)

Major topics

Trace

We have introduced a new component which replaces stopwatch called tracer. The core functionality remained the same, it still allows to measure time duration between two parts of the code and store it internally. The motive was that the original implementation was somehow cumbersome, and the name did not reflect the main purpose. New tracer object is created via constructor NewTracer(<name>, <logger>). It has very simple API with method LogTime(<name>, <start_time>) which adds entry to the internal database and Get(), which returns a proto-modelled list of entries. Database can be purged with Clear().

REST Security

Basic security support was added to RPC plugin. The caller can use pre-defined credentials to obtain authentication tokens required for access. Permission groups are also supported. To create a new permission group, use method from REST API RegisterPermissionGroup(<groups>). Group is assigned to the user in http.conf file (where all users are defined). This feature is available only to REST and it is considered experimental in the current release, and will be extended in the future.

New Features

  • measure
    • New component tracer was introduced. It serves the similar purpose as the stopwatch. More details in the readme
    • Stopwatch was removed
  • statuscheck
    • The liveness probe now shows also a state of all registered plugins (not only the overall state)
  • rest
    • New security functionality for REST plugin was added. To learn more about it, see the readme.
  • logging
    • Logger API has two new methods, SetOutput(<io.Writer>) to set custom logging output and SetFormatter(<formatter>) to set custom formatter before logged to output.

Other

  • Every proto file is generated with the gogo/proto package.

Bugfix

  • GRPC "listen and serve" call was moved to after init, which prevents calling services which were not yet registered. Also fixed a bug causing occasional panic if GRPC plugin was disabled.
  • ETCD reconnect resync fixed

Release v1.5 (2018-08-24)

Major topics

Flavors redesign

This version finally introduces new concept of flavors, system which handles plugins and dependencies between them. We believe that the new flavors are much more convenient and easier to understand and use. New system widely uses principle of options. Application NewAgent(options...) takes a list of options as a parameter. The most important option is AllPlugins(plugins...) which allows to set a list of plugins to the agent. To add single plugin, use Plugins.... Another options can be used to set timeouts, exit signal in form of go channel or version.

Every plugin defines default plugin instance and other useful methods to, for example, set custom plugin dependency. Plugins have their options.go file with implementation.

Examples go hand in hand with new flavors, so all of them were updated. The change affects only example plugin initialization, the main purpose was left unchanged.

Cryptodata plugin

New cryptodata plugin was added, providing support for encrypting and decrypting arbitrary data using configured private keys and support for wrapping key/value bytes/proto broker to automatically decrypt any read value matching specified pattern.

BoldDB store

Our key/value databases have a new member, BoltDB. Plugin uses own configuration file to define path to BoltDB file, permission, etc. Bolt supports all standard cn-infra features, including resync.

New Features

  • agent
    • Agent uses new concept of flavors. Since all the examples moved to this new plugin management, package flavors was removed. Learn more about new flavors here
  • cryptodata
    • New plugin cryptodata contains implementation files for encryption/decryption support. In order to try the functionality, examples for library , plugin and proto-plugin were added.
  • boltDB
    • Added support for the BoltDB keyval database. There is also a new example for BoltDB plugin.
  • logging
    • Added support for external hooks

Release v1.4.1 (2018-07-23)

Bugfix

  • Fixed issue in Consul client that caused brokers to incorrectly trim prefixes and thus storing invalid revisions for resync.

Release v1.4 (2018-07-16)

Breaking Changes

  • Package etcdv3 was renamed to etcd. This change affects imports. Also pay attention to the configuration flag which can also be influenced by the change. Based on the change, flag for ETCD configuration file --etcdv3-config is now defined as --etcd-config.

New Features

  • Support for GRPC unix domain socket type. Socket types tcp, tcp4, tcp6, unix and unixpacket can be used with GRPC. Desired socket type and address/file can be specified via grpc configuration file (example here). More information here
  • Rest plugin security improved. Security features are the usage of client/server certificates (HTTPS) and basic HTTP authentication with username and password. Those features are disabled by default. Information about how to use it and example can be found here

Other

  • Example configuration files with description were added to every plugin which supports/uses them.

Bugfix

  • Fixed occasional failure of method deriving config files
  • Fixed multiple issues in logs-lib example (logger, HTTP usage)
  • To prevent incorrect values in subsequent changes, previous value of key should be correctly cleaned up if the resync was called outside of initialization phase.
  • Fixed the logger configuration file. All created loggers are correctly set with a log level according to the map in file. The default log level can be also set, but keep in mind that the environmental variable INITIAL_LOGLVL replaces the value from the config.

Release v1.3 (2018-05-24)

New Features

  • Automatic resync if ETCD was disconnected and connected again. The feature is disabled by default. See plugin directory to learn how to enable the feature.
  • Watch registration API now contains new method Registration() allowing to register new key to all adapters.
  • New plugin for Consul. See plugin directory for more information.
  • In-memory mapping method UpdateMetadata() now triggers events. Use IsUpdate() so see if the event comes from the update notification.

Bugfix

  • Transport for statuscheck plugin fixed
  • Fixed bug where watcher was closed after server restart if database was compacted

Release v1.2 (2018-03-22)

New Features

  • Added support for ETCD compacting. Information about how to use it can be found in the readme
  • Name-to-index mapping API was extended with new method 'Update'. The purpose of the method is to update metadata value under specific key without triggering events, so mapping entry can be kept up to date.

Bugfix

  • Fixed syncbase issue where delete request for a non-existing item used to trigger a change notification
  • Getting of previous value for ProtoWatchResp 'delete' event now returns correct data

Release v1.1 (2018-02-07)

Dependencies

  • Migrated from glide to dep

Prometheus

  • Introduced Prometheus plugin with examples

Release v1.0.8 (2018-01-22)

Kafka

  • Added support for Kafka TLS.

Logging

  • Logger config file now enables to set every logger to desired level or use default level for all loggers within plugins. For this purpose it is also possible to use environment variable INITITAL_LOGLVL.

Statuscheck

  • Readiness probe now allows to report interfaces' state to the proble output.

Dependencies

  • Sirupsen package is now lower-cased according to recommandations.

Release v1.0.7 (2017-11-14)

Agent, Flavors

Input arguments of core.NewAgent() were changed:

  • it is possible to call NewAgent without options: core.NewAgent(flavor)
  • you can pass options like this: core.NewAgent(flavor, core.WithTimeout(1* time.Second))
  • there is core.NewAgentDeprecated() for backward compatibility

This release contains utilities/options to avoid writing the new flavor go structures (Inject, Plugins methods) for simple customizations:

  • if you just expose the RPCs you can write 
    rpc.NewAgent(rpc.WithPlugins(func(flavor *rpc.FlavorRPC) []*core.NamedPlugin {
  return []*core.NamedPlugin{{"myplugin1", &MyPlugin{&flavor.GRPC},
                             {"myplugin2", &MyPlugin{&flavor.GRPC},}
}))
  • if you want to use one simple plugin (without any client or a server) you can write: 
    flavor := &local.FlavorLocal{}
core.NewAgent(core.Inject(flavor), core.WithPlugin("myplugin1", &MyPlugin{Deps: flavor.PluginInfraDeps("myplugin1")}))
  • if you want to combine multiple flavors to inject their plugins to new MyPlugin 
    loc := &local.FlavorLocal{}
rpcs := &rpc.FlavorRPC{FlavorLocal: loc}
cons := &connectors.AllConnectorsFlavor{FlavorLocal: loc}
core.NewAgent(core.Inject(rpcs, cons), core.WithPlugin("myplugin", &MyPlugin{Deps: Deps{&rpcs.GRPC, &cons.ETCD}}))

ETCD/Datasync

  • GetPrevValue enabled in the proto watcher API. Etcd-lib/watcher example was updated to show the added functionality.
  • Fixed datasync internal state after resync causing that the resynced data were handled as created after first modification.
  • Fixed issue where datasync plugin was stuck on close

Cassandra

Added TLS support

Release v1.0.6 (2017-10-30)

ETCD/Datasync

  • etcd new feature PutIfNotExists adds key-value pair if the key doesn't exist.
  • feature GetPrevValue() used to obtain previous value from key-value database was returned to API
  • watcher registration object has a new method to close single subscribed key. Key can be un-subscribed in runtime. See example usage in examples/datasync-plugin for more details

Documentation

  • improved documentation/code comments in datasync, config and core packages

Release v1.0.5 (2017-10-17)

Profiling

  • new logging/measure - time measurement utility to measure duration of function or any part of the code. Use NewStopwatch(name string, log logging.Logger) to create an instance of stopwatch with name, desired logger and table with measured time entries. See plugin folder for more information.

Kafka

  • proto_connection.go and bytes_connection.go consolidated, bytes_connection.go now mirrors all the functionality from proto_connection.go.
    • mux can create two types of connection, standard bytes connection and bytes manual connection. This enables to call only respective methods on them (to use manual partitioning, it is needed to create manual connection, etc.)
    • method ConsumeTopicOnPartition renamed to ConsumeTopic (similar naming as in the proto_connection.go). The rest of the signature is not changed.
  • post-init watcher enabled in bytes_connection.go api
  • added methods MarkOffset and CommitOffsets to both, proto and bytes connection. Automatic offset marking was removed
  • one instance of mux in kafka plugin
  • new field group-id can be added to kafka.conf. This value is used as a Group ID in order to set it manually. In case the value is not provided, the service label is used instead (just like before).

Release v1.0.4 (2017-09-25)

Documentation

  • Improved documentation of public APIs (comments)
  • Improved documentation of examples (comments, doc.go, README.md)
  • Underscore in example suffixes "_lib" and "_plugin" replaced with a dash

Health, status check & probes

  • status check is now registered also for Cassandra & Redis
  • new prometheus format probe support (in rpcflavor)

Profiling

  • Logging duration (etcd connection establishment, kafka connection establishment, resync)

Plugin Configuration

  • new examples/configs-plugin
  • new flag --config-dir=. (by default "." meaning current working directory)
  • configuration files can but not need to have absolute paths anymore (e.g. --kafka-config=kafka.conf)
  • if you put all configuration files (etcd.conf, kafka.conf etc.) in one directory agent will load them
  • if you want to disable configuration file just put empty value for a particular flag (e.g. --kafka-config)

Logging

  • logmanager plugin
    • new optional flag --logs-config=logs.conf (showcase in examples/logs-plugin)
    • this plugin is now part of LocalFlavor (see field Logs) & tries to load configuration
    • HTTP dependency is optional (if it is not set it just does not registers HTTP handlers)
  • logger name added in logger fields (possible to grep only certain logger - effectively by plugin)

Kafka

  • kafka.Plugin.Disabled() returned if there is no kafka.conf present
  • Connection in bytes_connection.go renamed to BytesConnection
  • kafka plugin initializes two multiplexers for dynamic mode (automatic partitions) and manual mode. Every multiplexer can create its own connection and provides access to different set of methods (publishing to partition, watching on partition/offset)
  • ProtoWatcher from API was changed - methods WatchPartition and StopWatchPartition were removed from the ProtoWatcher interface and added to newly created ProtoPartitionWatcher. There is also a new method under Mux interface - NewPartitionWatcher(subscriber) which returns ProtoPartitionWatcher instance that allows to call partition-related methods
  • Offset mark is done for hash/default-partitioned messages only. Manually partitioned message's offset is not marked.
  • It is possible to start kafka consumer on partition after kafka plugin initialization procedure. New example post-init-consumer was created to show the functionality
  • fixes inside Mux.NewSyncPublisher() & Mux.NewAsyncPublisher() related to previous partition changes
  • Known Issues:
    • More than one network connection to Kafka (multiple instances of MUX)
    • TODO Minimalistic examples & documentation for Kafka API will be improved in a later release.

Flavors

  • optionally GPRC server can be enabled in rpc flavor (removed in v1.5) using --grpc-port=9111 (or using config gprc.conf)
  • Flavor interface (removed in v1.5) now contains three methods: Plugins(), Inject(), LogRegistry() to standardize these methods over all flavors. Note, LogRegistry() is usually embedded using local flavor.

Release v1.0.3 (2017-09-08)

  • FlavorAllConnectors (removed in v1.5)
    • Inlined plugins: ETCD, Kafka, Redis, Cassandra
  • Kafka Partitions
    • Implemented new methods that allow to specify partitions & offset parameters:
      • publish: Mux.NewSyncPublisherToPartition() & Mux.NewAsyncPublisherToPartition()
      • watch: ProtoWatcher.WatchPartition()
    • Minimalistic examples & documentation for Kafka API will be improved in a later release.

Release v1.0.2 (2017-08-28)

Major Themes

The major themes for Release v1.0.2 are as follows:

  • Libraries (GO Lang packages) for connecting to Data Bases and Message Bus. Set of these libraries provides unified client API and configuration for:
  • Data Synchronization plugin for watching and writing data asynchronously; it is currently implemented only for the db/keyval API API. It facilitates reading of data during startup or after reconnection to a data store and then watching incremental changes.
  • Agent Core that provides plugin lifecycle management (initialization and graceful shutdown of plugins) is able to run different flavors (removed in v1.5) (reusable collection of plugins):
    • local flavor - a minimal collection of plugins:
    • RPC flavor - exposes REST API for all plugins, especially for:
      • statuscheck (RPCs probed from systems such as K8s)
      • logging (for changing log level at runtime remotely)
    • connector flavors:
      • Cassandra flavor
      • etcd flavor
      • Redis flavor
      • Kafka flavor
  • Examples:
  • Docker container-based development environment
  • Helpers:
    • IDX Map is a reusable thread-safe in memory data structure. This map is designed for sharing key value based data (lookup by primary & secondary indexes, plus watching individual changes). It is useful for:
      • implementing backend for plugin specific API shared by multiple plugins;
      • caching of a remote data store.
    • Config(config: helpers for loading plugin specific configuration.
    • Service Label: retrieval of a unique identifier for a CN-Infra based app.