chore(deps): update busybox docker tag to v1.36.1

Reviewed-on: #358

CHANGELOG.md

@@ -12,11 +12,19 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 * Metrics endpoint for Prometheus integration
 * Search in list of targets using API & CLI
 * Allow filtering targets by arch value
+* Configurable global timeout for builds
 
 ### Changed
 
 * Rewrote cron expression logic in C
 * Updated codebase to V commit after 0.3.3
+* Agents now use worker threads and no longer spawn a new thread for every
+  build
+
+### Fixed
+
+* Package upload now fails if TCP connection is closed before all bytes have
+  been received
 
 ### Removed
 

Dockerfile

@@ -29,7 +29,7 @@ RUN if [ -n "${CI_COMMIT_SHA}" ]; then \
     fi
 
 
-FROM busybox:1.35.0
+FROM busybox:1.36.1
 
 ENV PATH=/bin \
     VIETER_DATA_DIR=/data \

src/agent/daemon.v

@@ -20,11 +20,13 @@ struct AgentDaemon {
 	client client.Client
 mut:
 	images ImageManager
-	// Which builds are currently running; length is conf.max_concurrent_builds
-	builds []BuildConfig
 	// Atomic variables used to detect when a build has finished; length is
-	// conf.max_concurrent_builds
+	// conf.max_concurrent_builds. This approach is used as the difference
+	// between a recently finished build and an empty build slot is important
+	// for knowing whether the agent is currently "active".
 	atomics []u64
+	// Channel used to send builds to worker threads
+	build_channel chan BuildConfig
 }
 
 // agent_init initializes a new agent
@@ -34,8 +36,8 @@ fn agent_init(logger log.Log, conf Config) AgentDaemon {
 		client: client.new(conf.address, conf.api_key)
 		conf: conf
 		images: new_image_manager(conf.image_rebuild_frequency * 60)
-		builds: []BuildConfig{len: conf.max_concurrent_builds}
 		atomics: []u64{len: conf.max_concurrent_builds}
+		build_channel: chan BuildConfig{cap: conf.max_concurrent_builds}
 	}
 
 	return d
@@ -43,6 +45,11 @@ fn agent_init(logger log.Log, conf Config) AgentDaemon {
 
 // run starts the actual agent daemon. This function will run forever.
 pub fn (mut d AgentDaemon) run() {
+	// Spawn worker threads
+	for builder_index in 0 .. d.conf.max_concurrent_builds {
+		spawn d.builder_thread(d.build_channel, builder_index)
+	}
+
 	// This is just so that the very first time the loop is ran, the jobs are
 	// always polled
 	mut last_poll_time := time.now().add_seconds(-d.conf.polling_frequency)
@@ -107,10 +114,10 @@ pub fn (mut d AgentDaemon) run() {
 				// It's technically still possible that the build image is
 				// removed in the very short period between building the
 				// builder image and starting a build container with it. If
-				// this happens, faith really just didn't want you to do this
+				// this happens, fate really just didn't want you to do this
 				// build.
 
-				d.start_build(config)
+				d.build_channel <- config
 				running++
 			}
 		}
@@ -147,22 +154,6 @@ fn (mut d AgentDaemon) update_atomics() (int, int) {
 	return finished, empty
 }
 
-// start_build starts a build for the given BuildConfig.
-fn (mut d AgentDaemon) start_build(config BuildConfig) bool {
-	for i in 0 .. d.atomics.len {
-		if stdatomic.load_u64(&d.atomics[i]) == agent.build_empty {
-			stdatomic.store_u64(&d.atomics[i], agent.build_running)
-			d.builds[i] = config
-
-			spawn d.run_build(i, config)
-
-			return true
-		}
-	}
-
-	return false
-}
-
 // run_build actually starts the build process for a given target.
 fn (mut d AgentDaemon) run_build(build_index int, config BuildConfig) {
 	d.linfo('started build: ${config}')
@@ -195,3 +186,12 @@ fn (mut d AgentDaemon) run_build(build_index int, config BuildConfig) {
 
 	stdatomic.store_u64(&d.atomics[build_index], agent.build_done)
 }
+
+// builder_thread is a thread that constantly listens for builds to process
+fn (mut d AgentDaemon) builder_thread(ch chan BuildConfig, builder_index int) {
+	for {
+		build_config := <-ch or { break }
+
+		d.run_build(builder_index, build_config)
+	}
+}

src/build/build.v

@@ -94,8 +94,8 @@ pub:
 }
 
 // build_target builds the given target. Internally it calls `build_config`.
-pub fn build_target(address string, api_key string, base_image_id string, target &Target, force bool) !BuildResult {
+pub fn build_target(address string, api_key string, base_image_id string, target &Target, force bool, timeout int) !BuildResult {
-	config := target.as_build_config(base_image_id, force)
+	config := target.as_build_config(base_image_id, force, timeout)
 
 	return build_config(address, api_key, config)
 }
@@ -136,9 +136,17 @@ pub fn build_config(address string, api_key string, config BuildConfig) !BuildRe
 	dd.container_start(id)!
 
 	mut data := dd.container_inspect(id)!
+	start_time := time.now()
 
 	// This loop waits until the container has stopped, so we can remove it after
 	for data.state.running {
+		if time.now() - start_time > config.timeout * time.second {
+			dd.container_kill(id)!
+			dd.container_remove(id)!
+
+			return error('Build killed due to timeout (${config.timeout}s)')
+		}
+
 		time.sleep(1 * time.second)
 
 		data = dd.container_inspect(id)!

src/build/queue.v

@@ -33,6 +33,8 @@ pub struct BuildJobQueue {
 	default_schedule &cron.Expression
 	// Base image to use for targets without defined base image
 	default_base_image string
+	// After how many minutes a build should be forcefully cancelled
+	default_build_timeout int
 mut:
 	mutex shared util.Dummy
 	// For each architecture, a priority queue is tracked
@@ -44,10 +46,11 @@ mut:
 }
 
 // new_job_queue initializes a new job queue
-pub fn new_job_queue(default_schedule &cron.Expression, default_base_image string) BuildJobQueue {
+pub fn new_job_queue(default_schedule &cron.Expression, default_base_image string, default_build_timeout int) BuildJobQueue {
 	return BuildJobQueue{
 		default_schedule: unsafe { default_schedule }
 		default_base_image: default_base_image
+		default_build_timeout: default_build_timeout
 		invalidated: map[int]time.Time{}
 	}
 }
@@ -80,7 +83,7 @@ pub fn (mut q BuildJobQueue) insert(input InsertConfig) ! {
 		mut job := BuildJob{
 			created: time.now()
 			single: input.single
-			config: input.target.as_build_config(q.default_base_image, input.force)
+			config: input.target.as_build_config(q.default_base_image, input.force, q.default_build_timeout)
 		}
 
 		if !input.now {

src/console/targets/build.v

@@ -6,7 +6,7 @@ import os
 import build
 
 // build locally builds the target with the given id.
-fn build_target(conf Config, target_id int, force bool) ! {
+fn build_target(conf Config, target_id int, force bool, timeout int) ! {
 	c := client.new(conf.address, conf.api_key)
 	target := c.get_target(target_id)!
 
@@ -16,7 +16,7 @@ fn build_target(conf Config, target_id int, force bool) ! {
 	image_id := build.create_build_image(conf.base_image)!
 
 	println('Running build...')
-	res := build.build_target(conf.address, conf.api_key, image_id, target, force)!
+	res := build.build_target(conf.address, conf.api_key, image_id, target, force, timeout)!
 
 	println('Removing build image...')
 

src/console/targets/targets.v

@@ -232,6 +232,12 @@ pub fn cmd() cli.Command {
 						description: 'Architecture to schedule build for. Required when using -remote.'
 						flag: cli.FlagType.string
 					},
+					cli.Flag{
+						name: 'timeout'
+						description: 'After how many minutes to cancel the build. Only applies to local builds.'
+						flag: cli.FlagType.int
+						default_value: ['3600']
+					},
 				]
 				execute: fn (cmd cli.Command) ! {
 					config_file := cmd.flags.get_string('config-file')!
@@ -239,6 +245,7 @@ pub fn cmd() cli.Command {
 
 					remote := cmd.flags.get_bool('remote')!
 					force := cmd.flags.get_bool('force')!
+					timeout := cmd.flags.get_int('timeout')!
 					target_id := cmd.args[0].int()
 
 					if remote {
@@ -251,7 +258,7 @@ pub fn cmd() cli.Command {
 						c := client.new(conf_.address, conf_.api_key)
 						c.queue_job(target_id, arch, force)!
 					} else {
-						build_target(conf_, target_id, force)!
+						build_target(conf_, target_id, force, timeout)!
 					}
 				}
 			},

src/models/builds.v

@@ -10,9 +10,10 @@ pub:
 	repo       string
 	base_image string
 	force      bool
+	timeout    int
 }
 
 // str return a single-line string representation of a build log
 pub fn (c BuildConfig) str() string {
-	return '{ target: ${c.target_id}, kind: ${c.kind}, url: ${c.url}, branch: ${c.branch}, path: ${c.path}, repo: ${c.repo}, base_image: ${c.base_image}, force: ${c.force} }'
+	return '{ target: ${c.target_id}, kind: ${c.kind}, url: ${c.url}, branch: ${c.branch}, path: ${c.path}, repo: ${c.repo}, base_image: ${c.base_image}, force: ${c.force}, timeout: ${c.timeout} }'
 }

src/models/targets.v

@@ -54,7 +54,7 @@ pub fn (t &Target) str() string {
 
 // as_build_config converts a Target into a BuildConfig, given some extra
 // needed information.
-pub fn (t &Target) as_build_config(base_image string, force bool) BuildConfig {
+pub fn (t &Target) as_build_config(base_image string, force bool, timeout int) BuildConfig {
 	return BuildConfig{
 		target_id: t.id
 		kind: t.kind
@@ -64,6 +64,7 @@ pub fn (t &Target) as_build_config(base_image string, force bool) BuildConfig {
 		repo: t.repo
 		base_image: base_image
 		force: force
+		timeout: timeout
 	}
 }
 

src/server/cli.v

@@ -16,6 +16,7 @@ pub:
 	max_log_age           int    [empty_default]
 	log_removal_schedule  string = '0 0'
 	collect_metrics       bool   [empty_default]
+	default_build_timeout int = 3600
 }
 
 // cmd returns the cli submodule that handles starting the server

src/server/repo.v

@@ -68,7 +68,7 @@ fn (mut app App) put_package(repo_ string) web.Result {
 		mut sw := time.new_stopwatch(time.StopWatchOptions{ auto_start: true })
 
 		util.reader_to_file(mut app.reader, length.int(), pkg_path) or {
-			app.lwarn("Failed to upload '${pkg_path}'")
+			app.lwarn("Failed to upload '${pkg_path}': ${err.msg()}")
 
 			return app.status(.internal_server_error)
 		}

src/server/server.v

@@ -108,7 +108,7 @@ pub fn server(conf Config) ! {
 		repo: repo_
 		db: db
 		collector: collector
-		job_queue: build.new_job_queue(global_ce, conf.base_image)
+		job_queue: build.new_job_queue(global_ce, conf.base_image, conf.default_build_timeout)
 	}
 	app.init_job_queue() or {
 		util.exit_with_message(1, 'Failed to inialize job queue: ${err.msg()}')

src/util/stream.v

@@ -46,6 +46,10 @@ pub fn reader_to_file(mut reader io.BufferedReader, length int, path string) ! {
 			to_write = to_write - bytes_written
 		}
 	}
+
+	if bytes_left > 0 {
+		return error('Not all bytes were received.')
+	}
 }
 
 // match_array_in_array[T] returns how many elements of a2 overlap with a1. For