Merge branch 'f-ignored-states'

CHANGELOG.md

@@ -1,3 +1,14 @@
+## 0.3.0 (Sep 6, 2022)
+
+NOTES:
+* This version changes the default behaviour when counting the servers in the pool.
+Previously servers in state `ERROR` were ignored. If you want to have the same
+behaviour as previously that can be configured using the new `ignored_states`
+configuration option
+
+FEATURES:
+* Allow setting a `ignored_states` to ignore server in these states when counting
+
 ## 0.2.5 (Apr 11, 2022)
 
 BUG FIXES:

README.md

@@ -40,6 +40,7 @@ target "os-nova" {
 * `id_attribute` `(string: "")` - The nomad attribute to use that maps the nomad client to an OS Compute instance. By default `unique.platform.aws.hostname` is used and a previous search is needed
 to get the instance id using the instance name
 * `action_timeout` `(string: "")` - The timeout to use when performing create and delete actions over servers. This should be specified as a duration. The default vaule is 90s
+* `ignored_states` `(string: "")` - A comma-separated list of server states to be ignored. The complete list can be seen [here](https://docs.openstack.org/api-guide/compute/server_concepts.html)
 
 ### Policy Configuration
 

plugin/openstack.go

@@ -436,18 +436,22 @@ func (t *TargetPlugin) countServers(ctx context.Context, pool string) (int64, in
 		if err := servers.ExtractServersInto(page, &serverList); err != nil {
 			return false, err
 		}
-		// TODO: check other status of servers: https://docs.openstack.org/api-guide/compute/server_concepts.html
-		// add option to allow setting the status of the servers to be counted
-		// and another option to specify what to do with servers in ERROR state (or to allow counting them as well)
+
 		for _, v := range serverList {
-			if v.Status == "ERROR" {
+			if _, ok := t.ignoredStates[v.Status]; ok {
+				t.logger.Debug("Ignored server due to state", "id", v.ID, "state", v.Status)
 				continue
 			}
-			if v.Status == "ACTIVE" {
+			switch v.Status {
+			case "ACTIVE":
 				ready += 1
+			case "BUILD", "REBOOT", "HARD_REBOOT":
+				// normal state transition, don't log but we need to wait
+			default:
+				t.logger.Warn("Detected server in unexpected status", "id", v.ID, "state", v.Status)
 			}
-			azDist[v.AZ] = azDist[v.AZ] + 1
 
+			azDist[v.AZ] = azDist[v.AZ] + 1
 			remoteIDs = append(remoteIDs, idFn(v))
 			total += 1
 		}

plugin/plugin.go

@@ -3,6 +3,7 @@ package plugin
 import (
 	"context"
 	"fmt"
+	"strings"
 	"time"
 
 	"github.com/gophercloud/gophercloud"
@@ -49,6 +50,7 @@ const (
 
 	configKeyValueSeparator = "value_separator"
 	configKeyActionTimeout  = "action_timeout"
+	configKeyIgnoredStates  = "ignored_states"
 
 	configKeyStopFirst   = "stop_first"
 	configKeyForceDelete = "force_delete"
@@ -77,6 +79,7 @@ type TargetPlugin struct {
 	avZones       []string
 	cache         map[string]string
 	actionTimeout int // seconds
+	ignoredStates map[string]struct{}
 
 	// clusterUtils provides general cluster scaling utilities for querying the
 	// state of nodes pools and performing scaling tasks.
@@ -98,13 +101,9 @@ func (t *TargetPlugin) SetConfig(config map[string]string) error {
 	if err := t.setupOSClients(config); err != nil {
 		return err
 	}
-	t.actionTimeout = defaultActionTimeout
-	if timeout, ok := config[configKeyActionTimeout]; ok {
-		d, err := time.ParseDuration(timeout)
-		if err != nil {
-			return fmt.Errorf("failed to parse action_timeout: %v", err)
-		}
-		t.actionTimeout = int(d.Seconds())
+
+	if err := t.configurePlugin(config); err != nil {
+		return err
 	}
 
 	clusterUtils, err := scaleutils.NewClusterScaleUtils(nomad.ConfigFromNamespacedMap(config), t.logger)
@@ -120,6 +119,32 @@ func (t *TargetPlugin) SetConfig(config map[string]string) error {
 	return nil
 }
 
+func (t *TargetPlugin) configurePlugin(config map[string]string) error {
+	t.actionTimeout = defaultActionTimeout
+	if timeout, ok := config[configKeyActionTimeout]; ok {
+		d, err := time.ParseDuration(timeout)
+		if err != nil {
+			return fmt.Errorf("failed to parse action_timeout: %v", err)
+		}
+		t.actionTimeout = int(d.Seconds())
+	}
+
+	t.ignoredStates = make(map[string]struct{})
+	if states, ok := config[configKeyIgnoredStates]; ok && strings.TrimSpace(states) != "" {
+		stateList := strings.Split(strings.TrimSpace(states), ",")
+		for _, name := range stateList {
+			state := strings.ToUpper(name)
+			switch state {
+			case "ACTIVE", "BUILD", "REBOOT", "HARD_REBOOT":
+				return fmt.Errorf("error setting ignored_states: state '%s' can't be ignored", state)
+			}
+			t.ignoredStates[state] = struct{}{}
+		}
+	}
+
+	return nil
+}
+
 // PluginInfo satisfies the PluginInfo function on the base.Base interface.
 func (t *TargetPlugin) PluginInfo() (*base.PluginInfo, error) {
 	return pluginInfo, nil