[Bugfix] [Relelase] 1.2.14-preview-1 (#1054)

Author: ajanikow

pkg/apis/deployment/v1/server_group.go

@@ -107,6 +107,13 @@ var (
 		ServerGroupSyncMasters,
 		ServerGroupSyncWorkers,
 	}
+	// AllArangoDServerGroups contains a constant list of all ArangoD server groups
+	AllArangoDServerGroups = []ServerGroup{
+		ServerGroupAgents,
+		ServerGroupSingle,
+		ServerGroupDBServers,
+		ServerGroupCoordinators,
+	}
 )
 
 // AsRole returns the "role" value for the given group.

pkg/apis/deployment/v2alpha1/server_group.go

@@ -107,6 +107,13 @@ var (
 		ServerGroupSyncMasters,
 		ServerGroupSyncWorkers,
 	}
+	// AllArangoDServerGroups contains a constant list of all ArangoD server groups
+	AllArangoDServerGroups = []ServerGroup{
+		ServerGroupAgents,
+		ServerGroupSingle,
+		ServerGroupDBServers,
+		ServerGroupCoordinators,
+	}
 )
 
 // AsRole returns the "role" value for the given group.

pkg/deployment/access_package.go

@@ -25,7 +25,7 @@ import (
 	"strings"
 	"time"
 
-	"gopkg.in/yaml.v3"
+	"gopkg.in/yaml.v2"
 	core "k8s.io/api/core/v1"
 	meta "k8s.io/apimachinery/pkg/apis/meta/v1"
 

pkg/deployment/agency/cache.go

@@ -24,11 +24,14 @@ import (
 	"context"
 	"sync"
 
+	"github.com/rs/zerolog"
+
 	"github.com/arangodb/go-driver"
 	"github.com/arangodb/go-driver/agency"
 
 	api "github.com/arangodb/kube-arangodb/pkg/apis/deployment/v1"
 	"github.com/arangodb/kube-arangodb/pkg/generated/metric_descriptions"
+	"github.com/arangodb/kube-arangodb/pkg/logging"
 	"github.com/arangodb/kube-arangodb/pkg/util/errors"
 	"github.com/arangodb/kube-arangodb/pkg/util/globals"
 	"github.com/arangodb/kube-arangodb/pkg/util/metrics"
@@ -164,10 +167,14 @@ func NewCache(namespace, name string, mode *api.DeploymentMode) Cache {
 }
 
 func NewAgencyCache(namespace, name string) Cache {
-	return &cache{
+	c := &cache{
 		namespace: namespace,
 		name:      name,
 	}
+
+	c.log = logger.WrapObj(c)
+
+	return c
 }
 
 func NewSingleCache() Cache {
@@ -197,6 +204,8 @@ func (c cacheSingle) Data() (State, bool) {
 type cache struct {
 	namespace, name string
 
+	log logging.Logger
+
 	lock sync.RWMutex
 
 	valid bool
@@ -208,6 +217,10 @@ type cache struct {
 	health Health
 }
 
+func (c *cache) WrapLogger(in *zerolog.Event) *zerolog.Event {
+	return in.Str("namespace", c.namespace).Str("name", c.name)
+}
+
 func (c *cache) CommitIndex() uint64 {
 	c.lock.RLock()
 	defer c.lock.RUnlock()
@@ -238,7 +251,7 @@ func (c *cache) Reload(ctx context.Context, size int, clients map[string]agency.
 	c.lock.Lock()
 	defer c.lock.Unlock()
 
-	leaderCli, leaderConfig, health, err := getLeader(ctx, size, clients)
+	leaderCli, leaderConfig, health, err := c.getLeader(ctx, size, clients)
 	if err != nil {
 		// Invalidate a leader ID and agency state.
 		// In the next iteration leaderID will be sat because `valid` will be false.
@@ -258,7 +271,7 @@ func (c *cache) Reload(ctx context.Context, size int, clients map[string]agency.
 	}
 
 	// A leader should be known even if an agency state is invalid.
-	if data, err := loadState(ctx, leaderCli); err != nil {
+	if data, err := c.loadState(ctx, leaderCli); err != nil {
 		c.valid = false
 		return leaderConfig.CommitIndex, err
 	} else {
@@ -271,7 +284,7 @@ func (c *cache) Reload(ctx context.Context, size int, clients map[string]agency.
 
 // getLeader returns config and client to a leader agency, and health to check if agencies are on the same page.
 // If there is no quorum for the leader then error is returned.
-func getLeader(ctx context.Context, size int, clients map[string]agency.Agency) (agency.Agency, *Config, health, error) {
+func (c *cache) getLeader(ctx context.Context, size int, clients map[string]agency.Agency) (agency.Agency, *Config, health, error) {
 	configs := make([]*Config, len(clients))
 	errs := make([]error, len(clients))
 	names := make([]string, 0, len(clients))
@@ -324,11 +337,11 @@ func getLeader(ctx context.Context, size int, clients map[string]agency.Agency)
 		}
 	}
 	if err := h.Serving(); err != nil {
-		logger.Err(err).Warn("Agency Not serving")
+		c.log.Err(err).Warn("Agency Not serving")
 		return nil, nil, h, err
 	}
 	if err := h.Healthy(); err != nil {
-		logger.Err(err).Warn("Agency Not healthy")
+		c.log.Err(err).Debug("Agency Not healthy")
 	}
 
 	for id := range names {

pkg/deployment/agency/state.go

@@ -31,7 +31,7 @@ import (
 	"github.com/arangodb/kube-arangodb/pkg/util/errors"
 )
 
-func loadState(ctx context.Context, client agency.Agency) (State, error) {
+func (c *cache) loadState(ctx context.Context, client agency.Agency) (State, error) {
 	conn := client.Connection()
 
 	req, err := client.Connection().NewRequest(http.MethodPost, "/_api/agency/read")
@@ -67,25 +67,17 @@ func loadState(ctx context.Context, client agency.Agency) (State, error) {
 		return State{}, err
 	}
 
-	var c StateRoots
+	var r StateRoots
 
-	if err := json.Unmarshal(data, &c); err != nil {
+	if err := json.Unmarshal(data, &r); err != nil {
 		return State{}, err
 	}
 
-	if len(c) != 1 {
+	if len(r) != 1 {
 		return State{}, errors.Newf("Invalid response size")
 	}
 
-	state := c[0].Arango
-
-	if _, ok := state.Current.Collections["_system"]; !ok {
-		return State{}, errors.Newf("Unable to find system database (invalid data)")
-	}
-
-	if _, ok := state.Plan.Collections["_system"]; !ok {
-		return State{}, errors.Newf("Unable to find system database (invalid data)")
-	}
+	state := r[0].Arango
 
 	return state, nil
 }

pkg/deployment/reconcile/action_backup_restore.go

@@ -22,6 +22,7 @@ package reconcile
 
 import (
 	"context"
+	"time"
 
 	"github.com/arangodb/go-driver"
 
@@ -203,6 +204,16 @@ func (a actionBackupRestore) CheckProgress(ctx context.Context) (bool, bool, err
 			// Retry
 			return false, false, nil
 		}
+
+		// Add wait grace period for restore jobs - async job creation is asynchronous
+		if ok := conn.IsAsyncErrorNotFound(restoreError); ok {
+			if s := a.action.StartTime; s != nil && !s.Time.IsZero() {
+				if time.Since(s.Time) < 10*time.Second {
+					// Retry
+					return false, false, nil
+				}
+			}
+		}
 	}
 
 	// Restore is done
@@ -214,6 +225,7 @@ func (a actionBackupRestore) CheckProgress(ctx context.Context) (bool, bool, err
 		}
 
 		if restoreError != nil {
+			a.log.Err(restoreError).Error("Restore failed")
 			result.State = api.DeploymentRestoreStateRestoreFailed
 			result.Message = restoreError.Error()
 		}

pkg/deployment/reconcile/plan_builder_tls.go

@@ -292,7 +292,7 @@ func (r *Reconciler) createKeyfileRenewalPlanSynced(ctx context.Context, apiObje
 		member := statusMember.Member
 
 		if !plan.IsEmpty() {
-			return nil
+			continue
 		}
 
 		cache, ok := planCtx.ACS().ClusterCache(member.ClusterID)