Replace COMPLEMENT_VERSION_CHECK_ITERATIONS with COMPLEMENT_SPAWN_HS_TIMEOUT_SECS (#234)

The new name is clearer and easier to understand. Fixes #232.

Currently, Complement will still accept the old env var but will log a Deprecated
warning. Homerunner will just look for the new env var HOMERUNNER_SPAWN_HS_TIMEOUT_SECS
and will ignore the old env var.

Author: kegsay

cmd/homerunner/README.md

@@ -4,9 +4,9 @@ An HTTP server which can spin up homeservers and execute blueprints on them. Pow
 
 Environment variables (all are optional):
 ```
-HOMERUNNER_LIFETIME_MINS=30                                       # how long networks can exist for before being destroyed
+HOMERUNNER_LIFETIME_MINS=30                                       # how long networks can exist for before being destroyed
 HOMERUNNER_PORT=54321                                             # port to listen on
-HOMERUNNER_VER_CHECK_ITERATIONS=100                               # how long to wait for the base image to spin up
+HOMERUNNER_SPAWN_HS_TIMEOUT_SECS=5                                # how long to wait for the base image to spin up
 HOMERUNNER_KEEP_BLUEPRINTS='clean_hs federation_one_to_one_room'  # space delimited blueprint names to keep images for
 HOMERUNNER_SNAPSHOT_BLUEPRINT=/some/file.json                     # single shot execute this blueprint then commit the image, does not run the server
 ```

cmd/homerunner/main.go

@@ -20,7 +20,7 @@ const Pkg = "homerunner"
 type Config struct {
 	HomeserverLifetimeMins int
 	Port                   int
-	VersionCheckIterations int
+	SpawnHSTimeout         time.Duration
 	KeepBlueprints         []string
 	Snapshot               string
 }
@@ -29,7 +29,7 @@ func NewConfig() *Config {
 	cfg := &Config{
 		HomeserverLifetimeMins: 30,
 		Port:                   54321,
-		VersionCheckIterations: 100,
+		SpawnHSTimeout:         5 * time.Second,
 		KeepBlueprints:         strings.Split(os.Getenv("HOMERUNNER_KEEP_BLUEPRINTS"), " "),
 		Snapshot:               os.Getenv("HOMERUNNER_SNAPSHOT_BLUEPRINT"),
 	}
@@ -39,20 +39,20 @@ func NewConfig() *Config {
 	if val, _ := strconv.Atoi(os.Getenv("HOMERUNNER_PORT")); val != 0 {
 		cfg.Port = val
 	}
-	if val, _ := strconv.Atoi(os.Getenv("HOMERUNNER_VER_CHECK_ITERATIONS")); val != 0 {
-		cfg.VersionCheckIterations = val
+	if val, _ := strconv.Atoi(os.Getenv("HOMERUNNER_SPAWN_HS_TIMEOUT_SECS")); val != 0 {
+		cfg.SpawnHSTimeout = time.Duration(val) * time.Second
 	}
 	return cfg
 }
 
 func cleanup(c *Config) {
 	cfg := &config.Complement{
-		PackageNamespace:       Pkg,
-		BaseImageURI:           "nothing",
-		DebugLoggingEnabled:    true,
-		VersionCheckIterations: c.VersionCheckIterations,
-		KeepBlueprints:         c.KeepBlueprints,
-		BestEffort:             true,
+		PackageNamespace:    Pkg,
+		BaseImageURI:        "nothing",
+		DebugLoggingEnabled: true,
+		SpawnHSTimeout:      c.SpawnHSTimeout,
+		KeepBlueprints:      c.KeepBlueprints,
+		BestEffort:          true,
 	}
 	builder, err := docker.NewBuilder(cfg)
 	if err != nil {

cmd/homerunner/setup.go

@@ -37,11 +37,11 @@ func (r *Runtime) CreateDeployment(imageURI string, blueprint *b.Blueprint) (*do
 	}
 	namespace := "homerunner_" + blueprint.Name
 	cfg := &config.Complement{
-		BaseImageURI:           imageURI,
-		DebugLoggingEnabled:    true,
-		VersionCheckIterations: r.Config.VersionCheckIterations,
-		BestEffort:             true,
-		PackageNamespace:       Pkg,
+		BaseImageURI:        imageURI,
+		DebugLoggingEnabled: true,
+		SpawnHSTimeout:      r.Config.SpawnHSTimeout,
+		BestEffort:          true,
+		PackageNamespace:    Pkg,
 	}
 	builder, err := docker.NewBuilder(cfg)
 	if err != nil {

internal/config/config.go

@@ -1,19 +1,21 @@
 package config
 
 import (
+	"fmt"
 	"os"
 	"strconv"
 	"strings"
+	"time"
 )
 
 type Complement struct {
-	BaseImageURI           string
-	BaseImageArgs          []string
-	DebugLoggingEnabled    bool
-	AlwaysPrintServerLogs  bool
-	BestEffort             bool
-	VersionCheckIterations int
-	KeepBlueprints         []string
+	BaseImageURI          string
+	BaseImageArgs         []string
+	DebugLoggingEnabled   bool
+	AlwaysPrintServerLogs bool
+	BestEffort            bool
+	SpawnHSTimeout        time.Duration
+	KeepBlueprints        []string
 	// The namespace for all complement created blueprints and deployments
 	PackageNamespace string
 }
@@ -24,7 +26,12 @@ func NewConfigFromEnvVars() *Complement {
 	cfg.BaseImageArgs = strings.Split(os.Getenv("COMPLEMENT_BASE_IMAGE_ARGS"), " ")
 	cfg.DebugLoggingEnabled = os.Getenv("COMPLEMENT_DEBUG") == "1"
 	cfg.AlwaysPrintServerLogs = os.Getenv("COMPLEMENT_ALWAYS_PRINT_SERVER_LOGS") == "1"
-	cfg.VersionCheckIterations = parseEnvWithDefault("COMPLEMENT_VERSION_CHECK_ITERATIONS", 100)
+	cfg.SpawnHSTimeout = time.Duration(parseEnvWithDefault("COMPLEMENT_SPAWN_HS_TIMEOUT_SECS", 30)) * time.Second
+	if os.Getenv("COMPLEMENT_VERSION_CHECK_ITERATIONS") != "" {
+		fmt.Fprintln(os.Stderr, "Deprecated: COMPLEMENT_VERSION_CHECK_ITERATIONS will be removed in a later version. Use COMPLEMENT_SPAWN_HS_TIMEOUT_SECS instead which does the same thing and is clearer.")
+		// each iteration had a 50ms sleep between tries so the timeout is 50 * iteration ms
+		cfg.SpawnHSTimeout = time.Duration(50*parseEnvWithDefault("COMPLEMENT_VERSION_CHECK_ITERATIONS", 100)) * time.Millisecond
+	}
 	cfg.KeepBlueprints = strings.Split(os.Getenv("COMPLEMENT_KEEP_BLUEPRINTS"), " ")
 	if cfg.BaseImageURI == "" {
 		panic("COMPLEMENT_BASE_IMAGE must be set")

internal/docker/builder.go

@@ -383,7 +383,7 @@ func (d *Builder) deployBaseImage(blueprintName string, hs b.Homeserver, context
 	return deployImage(
 		d.Docker, d.Config.BaseImageURI, d.CSAPIPort, fmt.Sprintf("complement_%s", contextStr),
 		d.Config.PackageNamespace, blueprintName, hs.Name, asIDToRegistrationMap, contextStr,
-		networkID, d.Config.VersionCheckIterations,
+		networkID, d.Config.SpawnHSTimeout,
 	)
 }
 
@@ -491,7 +491,8 @@ func generateASRegistrationYaml(as b.ApplicationService) string {
 }
 
 func deployImage(
-	docker *client.Client, imageID string, csPort int, containerName, pkgNamespace, blueprintName, hsName string, asIDToRegistrationMap map[string]string, contextStr, networkID string, versionCheckIterations int,
+	docker *client.Client, imageID string, csPort int, containerName, pkgNamespace, blueprintName, hsName string,
+	asIDToRegistrationMap map[string]string, contextStr, networkID string, spawnHSTimeout time.Duration,
 ) (*HomeserverDeployment, error) {
 	ctx := context.Background()
 	var extraHosts []string
@@ -595,7 +596,12 @@ func deployImage(
 	versionsURL := fmt.Sprintf("%s/_matrix/client/versions", baseURL)
 	// hit /versions to check it is up
 	var lastErr error
-	for i := 0; i < versionCheckIterations; i++ {
+	stopTime := time.Now().Add(spawnHSTimeout)
+	for {
+		if time.Now().After(stopTime) {
+			lastErr = fmt.Errorf("timed out checking for homeserver to be up: %s", lastErr)
+			break
+		}
 		res, err := http.Get(versionsURL)
 		if err != nil {
 			lastErr = fmt.Errorf("GET %s => error: %s", versionsURL, err)

internal/docker/deployer.go

@@ -89,7 +89,7 @@ func (d *Deployer) Deploy(ctx context.Context, blueprintName string) (*Deploymen
 		// TODO: Make CSAPI port configurable
 		deployment, err := deployImage(
 			d.Docker, img.ID, 8008, fmt.Sprintf("complement_%s_%s_%s_%d", d.config.PackageNamespace, d.DeployNamespace, contextStr, d.Counter),
-			d.config.PackageNamespace, blueprintName, hsName, asIDToRegistrationMap, contextStr, networkID, d.config.VersionCheckIterations)
+			d.config.PackageNamespace, blueprintName, hsName, asIDToRegistrationMap, contextStr, networkID, d.config.SpawnHSTimeout)
 		if err != nil {
 			if deployment != nil && deployment.ContainerID != "" {
 				// print logs to help debug