Merge pull request #258 from hashicorp/f-gh-257

scaleutils: add post scale in task to optionally purge nodes.

go.mod

@@ -16,7 +16,7 @@ require (
 	github.com/hashicorp/go-multierror v1.0.0
 	github.com/hashicorp/go-plugin v1.0.1
 	github.com/hashicorp/hcl/v2 v2.3.0
-	github.com/hashicorp/nomad/api v0.0.0-20200812215312-956c3a426dbc
+	github.com/hashicorp/nomad/api v0.0.0-20200904210342-cfe4f8314ff7
 	github.com/kr/pretty v0.2.0 // indirect
 	github.com/mattn/go-isatty v0.0.12 // indirect
 	github.com/mitchellh/cli v1.0.0

go.sum

@@ -118,8 +118,8 @@ github.com/hashicorp/golang-lru v0.5.0 h1:CL2msUPvZTLb5O648aiLNJw3hnBxN2+1Jq8rCO
 github.com/hashicorp/golang-lru v0.5.0/go.mod h1:/m3WP610KZHVQ1SGc6re/UDhFvYD7pJ4Ao+sR/qLZy8=
 github.com/hashicorp/hcl/v2 v2.3.0 h1:iRly8YaMwTBAKhn1Ybk7VSdzbnopghktCD031P8ggUE=
 github.com/hashicorp/hcl/v2 v2.3.0/go.mod h1:d+FwDBbOLvpAM3Z6J7gPj/VoAGkNe/gm352ZhjJ/Zv8=
-github.com/hashicorp/nomad/api v0.0.0-20200812215312-956c3a426dbc h1:b3Q1n+vuKmCLOa1H6cuet8xJT5LDaMiGJet55KnxW6w=
-github.com/hashicorp/nomad/api v0.0.0-20200812215312-956c3a426dbc/go.mod h1:DCi2k47yuUDzf2qWAK8E1RVmWgz/lc0jZQeEnICTxmY=
+github.com/hashicorp/nomad/api v0.0.0-20200904210342-cfe4f8314ff7 h1:xVUfoEZ6PAu+NZPZY/aRlG5FG/VosFS6kw1FjjEy30A=
+github.com/hashicorp/nomad/api v0.0.0-20200904210342-cfe4f8314ff7/go.mod h1:DCi2k47yuUDzf2qWAK8E1RVmWgz/lc0jZQeEnICTxmY=
 github.com/hashicorp/yamux v0.0.0-20180604194846-3520598351bb h1:b5rjCoWHc7eqmAS4/qyk21ZsHyb6Mxv/jykxvNTkU4M=
 github.com/hashicorp/yamux v0.0.0-20180604194846-3520598351bb/go.mod h1:+NfK9FKeTrX5uv1uIXGdwYDTeHna2qgaIlx54MXqjAM=
 github.com/jmespath/go-jmespath v0.0.0-20180206201540-c2b33e8439af h1:pmfjZENx5imkbgOkpRUYLnmbU7UEFbjtDA2hxJ1ichM=

helper/scaleutils/nomad.go

@@ -5,11 +5,13 @@ import (
 	"errors"
 	"fmt"
 	"os"
+	"strconv"
 	"sync"
 	"time"
 
 	hclog "github.com/hashicorp/go-hclog"
 	multierror "github.com/hashicorp/go-multierror"
+	"github.com/hashicorp/nomad-autoscaler/plugins/target"
 	"github.com/hashicorp/nomad/api"
 )
 
@@ -83,6 +85,53 @@ func (si *ScaleIn) RunPreScaleInTasks(ctx context.Context, req *ScaleInReq) ([]N
 	return nodeIDMap, nil
 }
 
+// RunPostScaleInTasks runs any tasks that need to occur after a remote node
+// provider has completed its work. It handles any users configuration so that
+// the plugin does not need to perform this work.
+func (si *ScaleIn) RunPostScaleInTasks(cfg map[string]string, nodes []NodeID) error {
+
+	// Attempt to read of the node purge config parameter. If it has been set
+	// then check its value, otherwise the default stance is that node purging
+	// is disabled.
+	if val, ok := cfg[target.ConfigKeyNodePurge]; ok {
+
+		// Parse the string as a bool. If we get an error return this as the
+		// operator has attempted to configure this value, but it's not worth
+		// breaking the whole pipeline for. Therefore log the error and return
+		// as Nomad will eventually perform this work.
+		boolVal, err := strconv.ParseBool(val)
+		if err != nil {
+			si.log.Error("failed to parse node_purge config param", "error", err)
+			return nil
+		}
+
+		// If the operator has disabled node purging, exit.
+		if !boolVal {
+			return nil
+		}
+	} else {
+		return nil
+	}
+
+	// Use a multierror to collect errors from any and all node purge calls
+	// that fail.
+	var mErr *multierror.Error
+
+	// Iterate the node list and perform a purge on each node. In the event of
+	// an error, add this to the list. Otherwise log useful information.
+	for _, node := range nodes {
+		resp, _, err := si.nomad.Nodes().Purge(node.NomadID, nil)
+		if err != nil {
+			mErr = multierror.Append(mErr, err)
+		} else {
+			si.log.Info("successfully purged Nomad node",
+				"node_id", node.NomadID, "nomad_evals", resp.EvalIDs)
+		}
+	}
+
+	return mErr.ErrorOrNil()
+}
+
 // identifyTargets filters the current Nomad cluster node list and then sorts
 // and selects nodes for removal based on the specified strategy. It is
 // possible the list does not contain as many nodes as requested. In this case,

plugins/builtin/target/aws-asg/plugin/aws.go

@@ -100,7 +100,7 @@ func (t *TargetPlugin) scaleIn(ctx context.Context, asg *autoscaling.AutoScaling
 
 	ids, err := t.scaleInUtils.RunPreScaleInTasks(ctx, scaleReq)
 	if err != nil {
-		return fmt.Errorf("failed to perform Nomad scale in tasks: %v", err)
+		return fmt.Errorf("failed to perform pre-scale Nomad scale in tasks: %v", err)
 	}
 
 	// Grab the instanceIDs once as it is used multiple times throughout the
@@ -139,6 +139,11 @@ func (t *TargetPlugin) scaleIn(ctx context.Context, asg *autoscaling.AutoScaling
 	log.Info("successfully terminated EC2 instances")
 	eWriter.write(ctx, scalingEventTerminate)
 
+	// Run any post scale in tasks that are desired.
+	if err := t.scaleInUtils.RunPostScaleInTasks(config, ids); err != nil {
+		return fmt.Errorf("failed to perform post-scale Nomad scale in tasks: %v", err)
+	}
+
 	return nil
 }
 

plugins/target/target.go

@@ -34,6 +34,7 @@ const (
 	ConfigKeyTaskGroup     = "Group"
 	ConfigKeyClass         = "node_class"
 	ConfigKeyDrainDeadline = "node_drain_deadline"
+	ConfigKeyNodePurge     = "node_purge"
 )
 
 // RPC is a plugin implementation that talks over net/rpc