Merge pull request #384 from ligato/dev

Release 2.0

CHANGELOG.md

@@ -1,3 +1,46 @@
+# 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](datasync/README.md)
+  - 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](idxmap/README.md)
+  - 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](rpc/grpc)
+  - Added authentication support for the GRPC plugin
+* [Probe](health/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](db/keyval/bolt)
+  - The config file now has an option to filter duplicated notifications.  
+
+## Improvements
+* [Agent](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](db/keyval/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](datasync/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

agent/agent.go

@@ -21,6 +21,7 @@ import (
 	"os/signal"
 	"runtime"
 	"strings"
+	"sync"
 	"time"
 
 	"github.com/ligato/cn-infra/config"
@@ -99,6 +100,9 @@ type agent struct {
 	stopOnce  once.ReturnError
 
 	tracer measure.Tracer
+
+	mu        sync.Mutex
+	curPlugin infra.Plugin
 }
 
 // Options returns the Options the agent was created with
@@ -144,10 +148,16 @@ func (a *agent) starter() error {
 	if timeout := a.opts.StartTimeout; timeout > 0 {
 		go func() {
 			select {
+			case s := <-sig:
+				agentLogger.Infof("Signal %v received during agent start, stopping", s)
+				os.Exit(1)
 			case <-started:
 				// agent started
 			case <-time.After(timeout):
-				agentLogger.Errorf("Agent failed to start before timeout (%v)", timeout)
+				a.mu.Lock()
+				curPlugin := a.curPlugin
+				a.mu.Unlock()
+				agentLogger.Errorf("Agent failed to start before timeout (%v) last plugin: %s", timeout, curPlugin)
 				dumpStacktrace()
 				os.Exit(1)
 			}
@@ -196,12 +206,16 @@ func (a *agent) starter() error {
 }
 
 func (a *agent) start() error {
-	agentLogger.Infof("Starting agent with %d plugins", len(a.opts.Plugins))
+	agentLogger.Debugf("starting %d plugins", len(a.opts.Plugins))
 
 	// Init plugins
 	for _, plugin := range a.opts.Plugins {
 		t := time.Now()
 
+		a.mu.Lock()
+		a.curPlugin = plugin
+		a.mu.Unlock()
+
 		agentLogger.Debugf("-> Init(): %v", plugin)
 		if err := plugin.Init(); err != nil {
 			return err
@@ -214,6 +228,10 @@ func (a *agent) start() error {
 	for _, plugin := range a.opts.Plugins {
 		t := time.Now()
 
+		a.mu.Lock()
+		a.curPlugin = plugin
+		a.mu.Unlock()
+
 		if postPlugin, ok := plugin.(infra.PostInit); ok {
 			agentLogger.Debugf("-> AfterInit(): %v", plugin)
 			if err := postPlugin.AfterInit(); err != nil {
@@ -226,6 +244,10 @@ func (a *agent) start() error {
 		a.tracer.LogTime(fmt.Sprintf("%v.AfterInit", plugin), t)
 	}
 
+	a.mu.Lock()
+	a.curPlugin = nil
+	a.mu.Unlock()
+
 	if printPluginStartDurations && infraLogger.GetLevel() >= logging.DebugLevel {
 		var b strings.Builder
 		b.WriteString("plugin start durations:\n")

agent/options.go

@@ -31,7 +31,7 @@ var (
 	DefaultStopTimeout = time.Second * 5
 
 	// DumpStackTraceOnTimeout prints stack trace on timeout or agent start/stop
-	DumpStackTraceOnTimeout = true
+	DumpStackTraceOnTimeout = os.Getenv("DUMP_STACK_ON_TIMEOUT") != ""
 )
 
 // Options specifies option list for the Agent

config/plugin_config.go

@@ -159,10 +159,7 @@ func Dir() (dir string, err error) {
 			if err != nil {
 				return cwd, err
 			}
-			if len(val) > 1 {
-				return filepath.Join(cwd, val[1:]), nil
-			}
-			return cwd, nil
+			return filepath.Join(cwd, val), nil
 		}
 		return val, nil
 	}

datasync/aggregator.go

@@ -16,6 +16,7 @@ package datasync
 
 import (
 	"github.com/gogo/protobuf/proto"
+	"github.com/ligato/cn-infra/logging"
 	"github.com/ligato/cn-infra/utils/safeclose"
 )
 
@@ -76,7 +77,9 @@ func (ta KVProtoWriters) Put(key string, data proto.Message, opts ...PutOption)
 // to add the key from that registration only
 func (wa *AggregatedRegistration) Register(resyncName, keyPrefix string) error {
 	for _, registration := range wa.Registrations {
-		registration.Register(resyncName, keyPrefix)
+		if err := registration.Register(resyncName, keyPrefix); err != nil {
+			logging.DefaultLogger.Warnf("aggregated register failed: %v", err)
+		}
 	}
 
 	return nil
@@ -86,7 +89,9 @@ func (wa *AggregatedRegistration) Register(resyncName, keyPrefix string) error {
 // Call Unregister(keyPrefix) on specific registration to remove the key from that registration only
 func (wa *AggregatedRegistration) Unregister(keyPrefix string) error {
 	for _, registration := range wa.Registrations {
-		registration.Unregister(keyPrefix)
+		if err := registration.Unregister(keyPrefix); err != nil {
+			logging.DefaultLogger.Warnf("aggregated unregister failed: %v", err)
+		}
 	}
 
 	return nil