diff --git a/src/Pure/Tools/build_job.scala b/src/Pure/Tools/build_job.scala --- a/src/Pure/Tools/build_job.scala +++ b/src/Pure/Tools/build_job.scala @@ -1,595 +1,590 @@ /* Title: Pure/Tools/build_job.scala Author: Makarius Build job running prover process, with rudimentary PIDE session. */ package isabelle import scala.collection.mutable trait Build_Job { def cancel(): Unit = () def is_finished: Boolean = false def join: (Process_Result, SHA1.Shasum) = (Process_Result.undefined, SHA1.no_shasum) } object Build_Job { - sealed case class Result(node_info: Host.Node_Info, process_result: Process_Result) { - def ok: Boolean = process_result.ok - } - - /* build session */ def start_session( build_context: Build_Process.Context, progress: Progress, log: Logger, session_background: Sessions.Background, input_shasum: SHA1.Shasum, node_info: Host.Node_Info ): Session_Job = { new Session_Job(build_context, progress, log, session_background, input_shasum, node_info) } object Session_Context { def load( build_uuid: String, name: String, deps: List[String], ancestors: List[String], session_prefs: String, sources_shasum: SHA1.Shasum, timeout: Time, store: Sessions.Store, progress: Progress = new Progress ): Session_Context = { def default: Session_Context = Session_Context( name, deps, ancestors, session_prefs, sources_shasum, timeout, Time.zero, Bytes.empty, build_uuid) store.try_open_database(name) match { case None => default case Some(db) => def ignore_error(msg: String) = { progress.echo_warning( "Ignoring bad database " + db + " for session " + quote(name) + if_proper(msg, ":\n" + msg)) default } try { val command_timings = store.read_command_timings(db, name) val elapsed = store.read_session_timing(db, name) match { case Markup.Elapsed(s) => Time.seconds(s) case _ => Time.zero } new Session_Context( name, deps, ancestors, session_prefs, sources_shasum, timeout, elapsed, command_timings, build_uuid) } catch { case ERROR(msg) => ignore_error(msg) case exn: java.lang.Error => ignore_error(Exn.message(exn)) case _: XML.Error => ignore_error("XML.Error") } finally { db.close() } } } } sealed case class Session_Context( name: String, deps: List[String], ancestors: List[String], session_prefs: String, sources_shasum: SHA1.Shasum, timeout: Time, old_time: Time, old_command_timings_blob: Bytes, build_uuid: String ) { override def toString: String = name } class Session_Job private[Build_Job]( build_context: Build_Process.Context, progress: Progress, log: Logger, session_background: Sessions.Background, input_shasum: SHA1.Shasum, node_info: Host.Node_Info ) extends Build_Job { private val store = build_context.store def session_name: String = session_background.session_name private val info: Sessions.Info = session_background.sessions_structure(session_name) private val options: Options = node_info.process_policy(info.options) private val session_sources = Sessions.Sources.load(session_background.base, cache = store.cache.compress) private val store_heap = build_context.store_heap(session_name) private val future_result: Future[(Process_Result, SHA1.Shasum)] = Future.thread("build", uninterruptible = true) { val env = Isabelle_System.settings( List("ISABELLE_ML_DEBUGGER" -> options.bool("ML_debugger").toString)) val session_heaps = session_background.info.parent match { case None => Nil case Some(logic) => ML_Process.session_heaps(store, session_background, logic = logic) } val use_prelude = if (session_heaps.isEmpty) Thy_Header.ml_roots.map(_._1) else Nil val eval_store = if (store_heap) { (if (info.theories.nonEmpty) List("ML_Heap.share_common_data ()") else Nil) ::: List("ML_Heap.save_child " + ML_Syntax.print_string_bytes(File.platform_path(store.output_heap(session_name)))) } else Nil def session_blobs(node_name: Document.Node.Name): List[(Command.Blob, Document.Blobs.Item)] = session_background.base.theory_load_commands.get(node_name.theory) match { case None => Nil case Some(spans) => val syntax = session_background.base.theory_syntax(node_name) val master_dir = Path.explode(node_name.master_dir) for (span <- spans; file <- span.loaded_files(syntax).files) yield { val src_path = Path.explode(file) val blob_name = Document.Node.Name(File.symbolic_path(master_dir + src_path)) val bytes = session_sources(blob_name.node).bytes val text = bytes.text val chunk = Symbol.Text_Chunk(text) Command.Blob(blob_name, src_path, Some((SHA1.digest(bytes), chunk))) -> Document.Blobs.Item(bytes, text, chunk, changed = false) } } /* session */ val resources = new Resources(session_background, log = log, command_timings = build_context.old_command_timings(session_name)) val session = new Session(options, resources) { override val cache: Term.Cache = store.cache override def build_blobs_info(node_name: Document.Node.Name): Command.Blobs_Info = Command.Blobs_Info.make(session_blobs(node_name)) override def build_blobs(node_name: Document.Node.Name): Document.Blobs = Document.Blobs.make(session_blobs(node_name)) } object Build_Session_Errors { private val promise: Promise[List[String]] = Future.promise def result: Exn.Result[List[String]] = promise.join_result def cancel(): Unit = promise.cancel() def apply(errs: List[String]): Unit = { try { promise.fulfill(errs) } catch { case _: IllegalStateException => } } } val export_consumer = Export.consumer(store.open_database(session_name, output = true), store.cache, progress = progress) val stdout = new StringBuilder(1000) val stderr = new StringBuilder(1000) val command_timings = new mutable.ListBuffer[Properties.T] val theory_timings = new mutable.ListBuffer[Properties.T] val session_timings = new mutable.ListBuffer[Properties.T] val runtime_statistics = new mutable.ListBuffer[Properties.T] val task_statistics = new mutable.ListBuffer[Properties.T] def fun( name: String, acc: mutable.ListBuffer[Properties.T], unapply: Properties.T => Option[Properties.T] ): (String, Session.Protocol_Function) = { name -> ((msg: Prover.Protocol_Output) => unapply(msg.properties) match { case Some(props) => acc += props; true case _ => false }) } session.init_protocol_handler(new Session.Protocol_Handler { override def exit(): Unit = Build_Session_Errors.cancel() private def build_session_finished(msg: Prover.Protocol_Output): Boolean = { val (rc, errors) = try { val (rc, errs) = { import XML.Decode._ pair(int, list(x => x))(Symbol.decode_yxml(msg.text)) } val errors = for (err <- errs) yield { val prt = Protocol_Message.expose_no_reports(err) Pretty.string_of(prt, metric = Symbol.Metric) } (rc, errors) } catch { case ERROR(err) => (Process_Result.RC.failure, List(err)) } session.protocol_command("Prover.stop", rc.toString) Build_Session_Errors(errors) true } private def loading_theory(msg: Prover.Protocol_Output): Boolean = msg.properties match { case Markup.Loading_Theory(Markup.Name(name)) => progress.theory(Progress.Theory(name, session = session_name)) false case _ => false } private def export_(msg: Prover.Protocol_Output): Boolean = msg.properties match { case Protocol.Export(args) => export_consumer.make_entry(session_name, args, msg.chunk) true case _ => false } override val functions: Session.Protocol_Functions = List( Markup.Build_Session_Finished.name -> build_session_finished, Markup.Loading_Theory.name -> loading_theory, Markup.EXPORT -> export_, fun(Markup.Theory_Timing.name, theory_timings, Markup.Theory_Timing.unapply), fun(Markup.Session_Timing.name, session_timings, Markup.Session_Timing.unapply), fun(Markup.Task_Statistics.name, task_statistics, Markup.Task_Statistics.unapply)) }) session.command_timings += Session.Consumer("command_timings") { case Session.Command_Timing(props) => for { elapsed <- Markup.Elapsed.unapply(props) elapsed_time = Time.seconds(elapsed) if elapsed_time.is_relevant && elapsed_time >= options.seconds("command_timing_threshold") } command_timings += props.filter(Markup.command_timing_property) } session.runtime_statistics += Session.Consumer("ML_statistics") { case Session.Runtime_Statistics(props) => runtime_statistics += props } session.finished_theories += Session.Consumer[Document.Snapshot]("finished_theories") { case snapshot => if (!progress.stopped) { def export_(name: String, xml: XML.Body, compress: Boolean = true): Unit = { if (!progress.stopped) { val theory_name = snapshot.node_name.theory val args = Protocol.Export.Args(theory_name = theory_name, name = name, compress = compress) val body = Bytes(Symbol.encode(YXML.string_of_body(xml))) export_consumer.make_entry(session_name, args, body) } } def export_text(name: String, text: String, compress: Boolean = true): Unit = export_(name, List(XML.Text(text)), compress = compress) for (command <- snapshot.snippet_command) { export_text(Export.DOCUMENT_ID, command.id.toString, compress = false) } export_text(Export.FILES, cat_lines(snapshot.node_files.map(name => File.symbolic_path(name.path))), compress = false) for ((blob_name, i) <- snapshot.node_files.tail.zipWithIndex) { val xml = snapshot.switch(blob_name).xml_markup() export_(Export.MARKUP + (i + 1), xml) } export_(Export.MARKUP, snapshot.xml_markup()) export_(Export.MESSAGES, snapshot.messages.map(_._1)) } } session.all_messages += Session.Consumer[Any]("build_session_output") { case msg: Prover.Output => val message = msg.message if (msg.is_system) resources.log(Protocol.message_text(message)) if (msg.is_stdout) { stdout ++= Symbol.encode(XML.content(message)) } else if (msg.is_stderr) { stderr ++= Symbol.encode(XML.content(message)) } else if (msg.is_exit) { val err = "Prover terminated" + (msg.properties match { case Markup.Process_Result(result) => ": " + result.print_rc case _ => "" }) Build_Session_Errors(List(err)) } case _ => } build_context.session_setup(session_name, session) val eval_main = Command_Line.ML_tool("Isabelle_Process.init_build ()" :: eval_store) /* process */ val process = Isabelle_Process.start(options, session, session_background, session_heaps, use_prelude = use_prelude, eval_main = eval_main, cwd = info.dir.file, env = env) val timeout_request: Option[Event_Timer.Request] = if (info.timeout_ignored) None else Some(Event_Timer.request(Time.now() + info.timeout) { process.terminate() }) val build_errors = Isabelle_Thread.interrupt_handler(_ => process.terminate()) { Exn.capture { process.await_startup() } match { case Exn.Res(_) => val resources_yxml = resources.init_session_yxml val encode_options: XML.Encode.T[Options] = options => session.prover_options(options).encode val args_yxml = YXML.string_of_body( { import XML.Encode._ pair(string, list(pair(encode_options, list(pair(string, properties)))))( (session_name, info.theories)) }) session.protocol_command("build_session", resources_yxml, args_yxml) Build_Session_Errors.result case Exn.Exn(exn) => Exn.Res(List(Exn.message(exn))) } } val result0 = Isabelle_Thread.interrupt_handler(_ => process.terminate()) { process.await_shutdown() } val was_timeout = timeout_request match { case None => false case Some(request) => !request.cancel() } session.stop() val export_errors = export_consumer.shutdown(close = true).map(Output.error_message_text) val (document_output, document_errors) = try { if (build_errors.isInstanceOf[Exn.Res[_]] && result0.ok && info.documents.nonEmpty) { using(Export.open_database_context(store)) { database_context => val documents = using(database_context.open_session(session_background)) { session_context => Document_Build.build_documents( Document_Build.context(session_context, progress = progress), output_sources = info.document_output, output_pdf = info.document_output) } using(database_context.open_database(session_name, output = true))(session_database => documents.foreach(_.write(session_database.db, session_name))) (documents.flatMap(_.log_lines), Nil) } } else (Nil, Nil) } catch { case exn: Document_Build.Build_Error => (exn.log_lines, exn.log_errors) case Exn.Interrupt.ERROR(msg) => (Nil, List(msg)) } /* process result */ val result1 = { val theory_timing = theory_timings.iterator.flatMap( { case props @ Markup.Name(name) => Some(name -> props) case _ => None }).toMap val used_theory_timings = for { (name, _) <- session_background.base.used_theories } yield theory_timing.getOrElse(name.theory, Markup.Name(name.theory)) val more_output = Library.trim_line(stdout.toString) :: command_timings.toList.map(Protocol.Command_Timing_Marker.apply) ::: used_theory_timings.map(Protocol.Theory_Timing_Marker.apply) ::: session_timings.toList.map(Protocol.Session_Timing_Marker.apply) ::: runtime_statistics.toList.map(Protocol.ML_Statistics_Marker.apply) ::: task_statistics.toList.map(Protocol.Task_Statistics_Marker.apply) ::: document_output result0.output(more_output) .error(Library.trim_line(stderr.toString)) .errors_rc(export_errors ::: document_errors) } val result2 = build_errors match { case Exn.Res(build_errs) => val errs = build_errs ::: document_errors if (errs.nonEmpty) { result1.error_rc.output( errs.flatMap(s => split_lines(Output.error_message_text(s))) ::: errs.map(Protocol.Error_Message_Marker.apply)) } else if (progress.stopped && result1.ok) result1.copy(rc = Process_Result.RC.interrupt) else result1 case Exn.Exn(Exn.Interrupt()) => if (result1.ok) result1.copy(rc = Process_Result.RC.interrupt) else result1 case Exn.Exn(exn) => throw exn } val process_result = if (result2.ok) result2 else if (was_timeout) result2.error(Output.error_message_text("Timeout")).timeout_rc else if (result2.interrupted) result2.error(Output.error_message_text("Interrupt")) else result2 /* output heap */ val output_shasum = if (process_result.ok && store_heap && store.output_heap(session_name).is_file) { SHA1.shasum(ML_Heap.write_digest(store.output_heap(session_name)), session_name) } else SHA1.no_shasum val log_lines = process_result.out_lines.filterNot(Protocol_Message.Marker.test) val build_log = Build_Log.Log_File(session_name, process_result.out_lines). parse_session_info( command_timings = true, theory_timings = true, ml_statistics = true, task_statistics = true) // write log file if (process_result.ok) { File.write_gzip(store.output_log_gz(session_name), terminate_lines(log_lines)) } else File.write(store.output_log(session_name), terminate_lines(log_lines)) // write database using(store.open_database(session_name, output = true))(db => store.write_session_info(db, session_name, session_sources, build_log = if (process_result.timeout) build_log.error("Timeout") else build_log, build = Sessions.Build_Info( sources = build_context.sources_shasum(session_name), input_heaps = input_shasum, output_heap = output_shasum, process_result.rc, build_context.build_uuid))) // messages process_result.err_lines.foreach(progress.echo(_)) if (process_result.ok) { val props = build_log.session_timing val threads = Markup.Session_Timing.Threads.unapply(props) getOrElse 1 val timing = Markup.Timing_Properties.get(props) progress.echo( "Timing " + session_name + " (" + threads + " threads, " + timing.message_factor + ")", verbose = true) progress.echo( "Finished " + session_name + " (" + process_result.timing.message_resources + ")") } else { progress.echo( session_name + " FAILED (see also \"isabelle build_log -H Error " + session_name + "\")") if (!process_result.interrupted) { val tail = info.options.int("process_output_tail") val suffix = if (tail == 0) log_lines else log_lines.drop(log_lines.length - tail max 0) val prefix = if (log_lines.length == suffix.length) Nil else List("...") progress.echo(Library.trim_line(cat_lines(prefix ::: suffix))) } } (process_result.copy(out_lines = log_lines), output_shasum) } override def cancel(): Unit = future_result.cancel() override def is_finished: Boolean = future_result.is_finished override def join: (Process_Result, SHA1.Shasum) = future_result.join } /* theory markup/messages from session database */ def read_theory( theory_context: Export.Theory_Context, unicode_symbols: Boolean = false ): Option[Document.Snapshot] = { def decode_bytes(bytes: Bytes): String = Symbol.output(unicode_symbols, UTF8.decode_permissive(bytes)) def read(name: String): Export.Entry = theory_context(name, permissive = true) def read_xml(name: String): XML.Body = YXML.parse_body(decode_bytes(read(name).bytes), cache = theory_context.cache) def read_source_file(name: String): Sessions.Source_File = theory_context.session_context.source_file(name) for { id <- theory_context.document_id() (thy_file, blobs_files) <- theory_context.files(permissive = true) } yield { val master_dir = Path.explode(Url.strip_base_name(thy_file).getOrElse( error("Cannot determine theory master directory: " + quote(thy_file)))) val blobs = blobs_files.map { name => val path = Path.explode(name) val src_path = File.relative_path(master_dir, path).getOrElse(path) val file = read_source_file(name) val bytes = file.bytes val text = decode_bytes(bytes) val chunk = Symbol.Text_Chunk(text) Command.Blob(Document.Node.Name(name), src_path, Some((file.digest, chunk))) -> Document.Blobs.Item(bytes, text, chunk, changed = false) } val thy_source = decode_bytes(read_source_file(thy_file).bytes) val thy_xml = read_xml(Export.MARKUP) val blobs_xml = for (i <- (1 to blobs.length).toList) yield read_xml(Export.MARKUP + i) val markups_index = Command.Markup_Index.make(blobs.map(_._1)) val markups = Command.Markups.make( for ((index, xml) <- markups_index.zip(thy_xml :: blobs_xml)) yield index -> Markup_Tree.from_XML(xml)) val results = Command.Results.make( for (elem @ XML.Elem(Markup(_, Markup.Serial(i)), _) <- read_xml(Export.MESSAGES)) yield i -> elem) val command = Command.unparsed(thy_source, theory = true, id = id, node_name = Document.Node.Name(thy_file, theory = theory_context.theory), blobs_info = Command.Blobs_Info.make(blobs), markups = markups, results = results) val doc_blobs = Document.Blobs.make(blobs) Document.State.init.snippet(command, doc_blobs) } } } diff --git a/src/Pure/Tools/build_process.scala b/src/Pure/Tools/build_process.scala --- a/src/Pure/Tools/build_process.scala +++ b/src/Pure/Tools/build_process.scala @@ -1,1092 +1,1116 @@ /* Title: Pure/Tools/build_process.scala Author: Makarius Build process for sessions, with build database, optional heap, and optional presentation. */ package isabelle import scala.collection.immutable.SortedMap import scala.math.Ordering import scala.annotation.tailrec object Build_Process { /** static context **/ object Context { def apply( store: Sessions.Store, build_deps: Sessions.Deps, progress: Progress = new Progress, ml_platform: String = Isabelle_System.getenv("ML_PLATFORM"), hostname: String = Isabelle_System.hostname(), numa_shuffling: Boolean = false, build_heap: Boolean = false, max_jobs: Int = 1, fresh_build: Boolean = false, no_build: Boolean = false, session_setup: (String, Session) => Unit = (_, _) => (), build_uuid: String = UUID.random().toString, master: Boolean = false, ): Context = { val sessions_structure = build_deps.sessions_structure val build_graph = sessions_structure.build_graph val sessions = Map.from( for ((name, (info, _)) <- build_graph.iterator) yield { val deps = info.parent.toList val ancestors = sessions_structure.build_requirements(deps) val sources_shasum = build_deps.sources_shasum(name) val session_context = Build_Job.Session_Context.load( build_uuid, name, deps, ancestors, info.session_prefs, sources_shasum, info.timeout, store, progress = progress) name -> session_context }) val sessions_time = { val maximals = build_graph.maximals.toSet def descendants_time(name: String): Double = { if (maximals.contains(name)) sessions(name).old_time.seconds else { val descendants = build_graph.all_succs(List(name)).toSet val g = build_graph.restrict(descendants) (0.0 :: g.maximals.flatMap { desc => val ps = g.all_preds(List(desc)) if (ps.exists(p => !sessions.isDefinedAt(p))) None else Some(ps.map(p => sessions(p).old_time.seconds).sum) }).max } } Map.from( for (name <- sessions.keysIterator) yield name -> descendants_time(name)).withDefaultValue(0.0) } val ordering = new Ordering[String] { def compare(name1: String, name2: String): Int = sessions_time(name2) compare sessions_time(name1) match { case 0 => sessions(name2).timeout compare sessions(name1).timeout match { case 0 => name1 compare name2 case ord => ord } case ord => ord } } val numa_nodes = Host.numa_nodes(enabled = numa_shuffling) new Context(store, build_deps, sessions, ordering, ml_platform, hostname, numa_nodes, build_heap = build_heap, max_jobs = max_jobs, fresh_build = fresh_build, no_build = no_build, session_setup, build_uuid = build_uuid, master = master) } } final class Context private( val store: Sessions.Store, val build_deps: Sessions.Deps, val sessions: State.Sessions, val ordering: Ordering[String], val ml_platform: String, val hostname: String, val numa_nodes: List[Int], val build_heap: Boolean, val max_jobs: Int, val fresh_build: Boolean, val no_build: Boolean, val session_setup: (String, Session) => Unit, val build_uuid: String, val master: Boolean ) { override def toString: String = "Build_Process.Context(build_uuid = " + quote(build_uuid) + if_proper(master, ", master = true") + ")" def build_options: Options = store.options def sessions_structure: Sessions.Structure = build_deps.sessions_structure def sources_shasum(name: String): SHA1.Shasum = sessions(name).sources_shasum def old_command_timings(name: String): List[Properties.T] = sessions.get(name) match { case Some(session_context) => Properties.uncompress(session_context.old_command_timings_blob, cache = store.cache) case None => Nil } def prepare_database(): Unit = { using_option(store.open_build_database()) { db => db.transaction { Data.all_tables.create_lock(db) Data.clean_build(db) } db.rebuild() } } def store_heap(name: String): Boolean = build_heap || Sessions.is_pure(name) || sessions.valuesIterator.exists(_.ancestors.contains(name)) def worker_active: Boolean = max_jobs > 0 } /** dynamic state **/ type Progress_Messages = SortedMap[Long, Progress.Message] case class Worker( worker_uuid: String, build_uuid: String, hostname: String, java_pid: Long, java_start: Date, start: Date, stamp: Date, stop: Option[Date], serial: Long ) case class Task( name: String, deps: List[String], info: JSON.Object.T = JSON.Object.empty ) { def is_ready: Boolean = deps.isEmpty def resolve(dep: String): Task = if (deps.contains(dep)) copy(deps = deps.filterNot(_ == dep)) else this } case class Job( name: String, worker_uuid: String, build_uuid: String, node_info: Host.Node_Info, build: Option[Build_Job] ) { def no_build: Job = copy(build = None) } case class Result( + name: String, + worker_uuid: String, + build_uuid: String, + node_info: Host.Node_Info, process_result: Process_Result, output_shasum: SHA1.Shasum, - node_info: Host.Node_Info, current: Boolean ) { def ok: Boolean = process_result.ok } object State { type Sessions = Map[String, Build_Job.Session_Context] type Workers = List[Worker] type Pending = List[Task] type Running = Map[String, Job] type Results = Map[String, Result] def inc_serial(serial: Long): Long = { require(serial < java.lang.Long.MAX_VALUE, "serial overflow") serial + 1 } } sealed case class State( serial: Long = 0, progress_seen: Long = 0, numa_next: Int = 0, sessions: State.Sessions = Map.empty, // static build targets workers: State.Workers = Nil, // available worker processes pending: State.Pending = Nil, // dynamic build "queue" running: State.Running = Map.empty, // presently running jobs results: State.Results = Map.empty // finished results ) { require(serial >= 0, "serial underflow") def inc_serial: State = copy(serial = State.inc_serial(serial)) def set_serial(i: Long): State = { require(serial <= i, "non-monotonic change of serial") copy(serial = i) } def progress_serial(message_serial: Long = serial): State = if (message_serial > progress_seen) copy(progress_seen = message_serial) else error("Bad serial " + message_serial + " for progress output (already seen)") def set_workers(new_workers: State.Workers): State = copy(workers = new_workers) def next_numa_node(numa_nodes: List[Int]): (Option[Int], State) = if (numa_nodes.isEmpty) (None, this) else { val available = numa_nodes.zipWithIndex val used = Set.from(for (job <- running.valuesIterator; i <- job.node_info.numa_node) yield i) val numa_index = available.collectFirst({ case (n, i) if n == numa_next => i }).getOrElse(0) val candidates = available.drop(numa_index) ::: available.take(numa_index) val (n, i) = candidates.find({ case (n, i) => i == numa_index && !used(n) }) orElse candidates.find({ case (n, _) => !used(n) }) getOrElse candidates.head (Some(n), copy(numa_next = numa_nodes((i + 1) % numa_nodes.length))) } def finished: Boolean = pending.isEmpty def remove_pending(name: String): State = copy(pending = pending.flatMap( entry => if (entry.name == name) None else Some(entry.resolve(name)))) def is_running(name: String): Boolean = running.isDefinedAt(name) def stop_running(): Unit = for (job <- running.valuesIterator; build <- job.build) build.cancel() def finished_running(): List[Job] = List.from( for (job <- running.valuesIterator; build <- job.build if build.is_finished) yield job) def add_running(job: Job): State = copy(running = running + (job.name -> job)) def remove_running(name: String): State = copy(running = running - name) def make_result( - name: String, + result_name: (String, String, String), process_result: Process_Result, output_shasum: SHA1.Shasum, node_info: Host.Node_Info = Host.Node_Info.none, current: Boolean = false ): State = { - val entry = name -> Build_Process.Result(process_result, output_shasum, node_info, current) - copy(results = results + entry) + val (name, worker_uuid, build_uuid) = result_name + val result = + Result(name, worker_uuid, build_uuid, node_info, process_result, output_shasum, current) + copy(results = results + (name -> result)) } } /** SQL data model **/ object Data { def make_table(name: String, columns: List[SQL.Column], body: String = ""): SQL.Table = SQL.Table("isabelle_build" + if_proper(name, "_" + name), columns, body = body) object Generic { val build_uuid = SQL.Column.string("build_uuid") val worker_uuid = SQL.Column.string("worker_uuid") val name = SQL.Column.string("name") def sql( build_uuid: String = "", worker_uuid: String = "", name: String = "", names: Iterable[String] = Nil ): SQL.Source = SQL.and( if_proper(build_uuid, Generic.build_uuid.equal(build_uuid)), if_proper(worker_uuid, Generic.worker_uuid.equal(worker_uuid)), if_proper(name, Generic.name.equal(name)), if_proper(names, Generic.name.member(names))) } /* base table */ object Base { val build_uuid = Generic.build_uuid.make_primary_key val ml_platform = SQL.Column.string("ml_platform") val options = SQL.Column.string("options") val start = SQL.Column.date("start") val stop = SQL.Column.date("stop") val progress_stopped = SQL.Column.bool("progress_stopped") val table = make_table("", List(build_uuid, ml_platform, options, start, stop, progress_stopped)) } def start_build( db: SQL.Database, build_uuid: String, ml_platform: String, options: String, progress_stopped: Boolean ): Unit = { db.execute_statement(Base.table.insert(), body = { stmt => stmt.string(1) = build_uuid stmt.string(2) = ml_platform stmt.string(3) = options stmt.date(4) = db.now() stmt.date(5) = None stmt.bool(6) = progress_stopped }) } def stop_build(db: SQL.Database, build_uuid: String): Unit = db.execute_statement( Base.table.update(List(Base.stop), sql = Base.build_uuid.where_equal(build_uuid)), body = { stmt => stmt.date(1) = db.now() }) def clean_build(db: SQL.Database): Unit = { val old = db.execute_query_statement( Base.table.select(List(Base.build_uuid), sql = SQL.where(Base.stop.defined)), List.from[String], res => res.string(Base.build_uuid)) if (old.nonEmpty) { for (table <- List(Base.table, Sessions.table, Progress.table, Workers.table)) { db.execute_statement(table.delete(sql = Generic.build_uuid.where_member(old))) } } } /* sessions */ object Sessions { val name = Generic.name.make_primary_key val deps = SQL.Column.string("deps") val ancestors = SQL.Column.string("ancestors") val options = SQL.Column.string("options") val sources = SQL.Column.string("sources") val timeout = SQL.Column.long("timeout") val old_time = SQL.Column.long("old_time") val old_command_timings = SQL.Column.bytes("old_command_timings") val build_uuid = Generic.build_uuid val table = make_table("sessions", List(name, deps, ancestors, options, sources, timeout, old_time, old_command_timings, build_uuid)) } def read_sessions_domain(db: SQL.Database): Set[String] = db.execute_query_statement( Sessions.table.select(List(Sessions.name)), Set.from[String], res => res.string(Sessions.name)) def read_sessions(db: SQL.Database, names: Iterable[String] = Nil): State.Sessions = db.execute_query_statement( Sessions.table.select(sql = if_proper(names, Sessions.name.where_member(names))), Map.from[String, Build_Job.Session_Context], { res => val name = res.string(Sessions.name) val deps = split_lines(res.string(Sessions.deps)) val ancestors = split_lines(res.string(Sessions.ancestors)) val options = res.string(Sessions.options) val sources_shasum = SHA1.fake_shasum(res.string(Sessions.sources)) val timeout = Time.ms(res.long(Sessions.timeout)) val old_time = Time.ms(res.long(Sessions.old_time)) val old_command_timings_blob = res.bytes(Sessions.old_command_timings) val build_uuid = res.string(Sessions.build_uuid) name -> Build_Job.Session_Context(name, deps, ancestors, options, sources_shasum, timeout, old_time, old_command_timings_blob, build_uuid) } ) def update_sessions(db:SQL.Database, sessions: State.Sessions): Boolean = { val old_sessions = read_sessions_domain(db) val insert = sessions.iterator.filterNot(p => old_sessions.contains(p._1)).toList for ((name, session) <- insert) { db.execute_statement(Sessions.table.insert(), body = { stmt => stmt.string(1) = name stmt.string(2) = cat_lines(session.deps) stmt.string(3) = cat_lines(session.ancestors) stmt.string(4) = session.session_prefs stmt.string(5) = session.sources_shasum.toString stmt.long(6) = session.timeout.ms stmt.long(7) = session.old_time.ms stmt.bytes(8) = session.old_command_timings_blob stmt.string(9) = session.build_uuid }) } insert.nonEmpty } /* progress */ object Progress { val serial = SQL.Column.long("serial").make_primary_key val kind = SQL.Column.int("kind") val text = SQL.Column.string("text") val verbose = SQL.Column.bool("verbose") val build_uuid = Generic.build_uuid val table = make_table("progress", List(serial, kind, text, verbose, build_uuid)) } def read_progress(db: SQL.Database, seen: Long = 0, build_uuid: String = ""): Progress_Messages = db.execute_query_statement( Progress.table.select( sql = SQL.where( SQL.and( if (seen <= 0) "" else Progress.serial.ident + " > " + seen, Generic.sql(build_uuid = build_uuid)))), SortedMap.from[Long, isabelle.Progress.Message], { res => val serial = res.long(Progress.serial) val kind = isabelle.Progress.Kind(res.int(Progress.kind)) val text = res.string(Progress.text) val verbose = res.bool(Progress.verbose) serial -> isabelle.Progress.Message(kind, text, verbose = verbose) } ) def write_progress( db: SQL.Database, message_serial: Long, message: isabelle.Progress.Message, build_uuid: String ): Unit = { db.execute_statement(Progress.table.insert(), body = { stmt => stmt.long(1) = message_serial stmt.int(2) = message.kind.id stmt.string(3) = message.text stmt.bool(4) = message.verbose stmt.string(5) = build_uuid }) } def sync_progress( db: SQL.Database, seen: Long, build_uuid: String, build_progress: Progress ): (Progress_Messages, Boolean) = { require(build_uuid.nonEmpty) val messages = read_progress(db, seen = seen, build_uuid = build_uuid) val stopped_db = db.execute_query_statementO[Boolean]( Base.table.select(List(Base.progress_stopped), sql = SQL.where(Base.build_uuid.equal(build_uuid))), res => res.bool(Base.progress_stopped) ).getOrElse(false) def stop_db(): Unit = db.execute_statement( Base.table.update( List(Base.progress_stopped), sql = Base.build_uuid.where_equal(build_uuid)), body = { stmt => stmt.bool(1) = true }) val stopped = build_progress.stopped if (stopped_db && !stopped) build_progress.stop() if (stopped && !stopped_db) stop_db() (messages, messages.isEmpty && stopped_db == stopped) } /* workers */ object Workers { val worker_uuid = Generic.worker_uuid.make_primary_key val build_uuid = Generic.build_uuid val hostname = SQL.Column.string("hostname") val java_pid = SQL.Column.long("java_pid") val java_start = SQL.Column.date("java_start") val start = SQL.Column.date("start") val stamp = SQL.Column.date("stamp") val stop = SQL.Column.date("stop") val serial = SQL.Column.long("serial") val table = make_table("workers", List(worker_uuid, build_uuid, hostname, java_pid, java_start, start, stamp, stop, serial)) val serial_max = serial.copy(expr = "MAX(" + serial.ident + ")") } def read_workers( db: SQL.Database, build_uuid: String = "", worker_uuid: String = "" ): State.Workers = { db.execute_query_statement( Workers.table.select(sql = SQL.where(Generic.sql(build_uuid = build_uuid, worker_uuid = worker_uuid))), List.from[Worker], { res => Worker( worker_uuid = res.string(Workers.worker_uuid), build_uuid = res.string(Workers.build_uuid), hostname = res.string(Workers.hostname), java_pid = res.long(Workers.java_pid), java_start = res.date(Workers.java_start), start = res.date(Workers.start), stamp = res.date(Workers.stamp), stop = res.get_date(Workers.stop), serial = res.long(Workers.serial)) }) } def serial_max(db: SQL.Database): Long = db.execute_query_statementO[Long]( Workers.table.select(List(Workers.serial_max)), res => res.long(Workers.serial) ).getOrElse(0L) def start_worker( db: SQL.Database, worker_uuid: String, build_uuid: String, hostname: String, java_pid: Long, java_start: Date ): Long = { def err(msg: String): Nothing = error("Cannot start worker " + worker_uuid + if_proper(msg, "\n" + msg)) val build_stop = db.execute_query_statementO( Base.table.select(List(Base.stop), sql = Base.build_uuid.where_equal(build_uuid)), res => res.get_date(Base.stop)) build_stop match { case Some(None) => case Some(Some(_)) => err("for already stopped build process " + build_uuid) case None => err("for unknown build process " + build_uuid) } val serial = serial_max(db) db.execute_statement(Workers.table.insert(), body = { stmt => val now = db.now() stmt.string(1) = worker_uuid stmt.string(2) = build_uuid stmt.string(3) = hostname stmt.long(4) = java_pid stmt.date(5) = java_start stmt.date(6) = now stmt.date(7) = now stmt.date(8) = None stmt.long(9) = serial }) serial } def stamp_worker( db: SQL.Database, worker_uuid: String, serial: Long, stop: Boolean = false ): Unit = { val sql = Workers.table.update(List(Workers.stamp, Workers.stop, Workers.serial), sql = Workers.worker_uuid.where_equal(worker_uuid)) db.execute_statement(sql, body = { stmt => val now = db.now() stmt.date(1) = now stmt.date(2) = if (stop) Some(now) else None stmt.long(3) = serial }) } /* pending jobs */ object Pending { val name = Generic.name.make_primary_key val deps = SQL.Column.string("deps") val info = SQL.Column.string("info") val table = make_table("pending", List(name, deps, info)) } def read_pending(db: SQL.Database): List[Task] = db.execute_query_statement( Pending.table.select(sql = SQL.order_by(List(Pending.name))), List.from[Task], { res => val name = res.string(Pending.name) val deps = res.string(Pending.deps) val info = res.string(Pending.info) Task(name, split_lines(deps), info = JSON.Object.parse(info)) }) def update_pending(db: SQL.Database, pending: State.Pending): Boolean = { val old_pending = read_pending(db) val (delete, insert) = Library.symmetric_difference(old_pending, pending) if (delete.nonEmpty) { db.execute_statement( Pending.table.delete(sql = SQL.where(Generic.sql(names = delete.map(_.name))))) } for (entry <- insert) { db.execute_statement(Pending.table.insert(), body = { stmt => stmt.string(1) = entry.name stmt.string(2) = cat_lines(entry.deps) stmt.string(3) = JSON.Format(entry.info) }) } delete.nonEmpty || insert.nonEmpty } /* running jobs */ object Running { val name = Generic.name.make_primary_key val worker_uuid = Generic.worker_uuid val build_uuid = Generic.build_uuid val hostname = SQL.Column.string("hostname") val numa_node = SQL.Column.int("numa_node") val table = make_table("running", List(name, worker_uuid, build_uuid, hostname, numa_node)) } def read_running(db: SQL.Database): List[Job] = db.execute_query_statement( Running.table.select(sql = SQL.order_by(List(Running.name))), List.from[Job], { res => val name = res.string(Running.name) val worker_uuid = res.string(Running.worker_uuid) val build_uuid = res.string(Running.build_uuid) val hostname = res.string(Running.hostname) val numa_node = res.get_int(Running.numa_node) Job(name, worker_uuid, build_uuid, Host.Node_Info(hostname, numa_node), None) } ) def update_running(db: SQL.Database, running: State.Running): Boolean = { val running0 = read_running(db) val running1 = running.valuesIterator.map(_.no_build).toList val (delete, insert) = Library.symmetric_difference(running0, running1) if (delete.nonEmpty) { db.execute_statement( Running.table.delete(sql = SQL.where(Generic.sql(names = delete.map(_.name))))) } for (job <- insert) { db.execute_statement(Running.table.insert(), body = { stmt => stmt.string(1) = job.name stmt.string(2) = job.worker_uuid stmt.string(3) = job.build_uuid stmt.string(4) = job.node_info.hostname stmt.int(5) = job.node_info.numa_node }) } delete.nonEmpty || insert.nonEmpty } /* job results */ object Results { val name = Generic.name.make_primary_key + val worker_uuid = Generic.worker_uuid + val build_uuid = Generic.build_uuid val hostname = SQL.Column.string("hostname") val numa_node = SQL.Column.string("numa_node") val rc = SQL.Column.int("rc") val out = SQL.Column.string("out") val err = SQL.Column.string("err") val timing_elapsed = SQL.Column.long("timing_elapsed") val timing_cpu = SQL.Column.long("timing_cpu") val timing_gc = SQL.Column.long("timing_gc") + val output_shasum = SQL.Column.string("output_shasum") + val current = SQL.Column.bool("current") val table = make_table("results", - List(name, hostname, numa_node, rc, out, err, timing_elapsed, timing_cpu, timing_gc)) + List(name, worker_uuid, build_uuid, hostname, numa_node, + rc, out, err, timing_elapsed, timing_cpu, timing_gc, output_shasum, current)) } def read_results_domain(db: SQL.Database): Set[String] = db.execute_query_statement( Results.table.select(List(Results.name)), Set.from[String], res => res.string(Results.name)) - def read_results(db: SQL.Database, names: List[String] = Nil): Map[String, Build_Job.Result] = + def read_results(db: SQL.Database, names: List[String] = Nil): State.Results = db.execute_query_statement( Results.table.select(sql = if_proper(names, Results.name.where_member(names))), - Map.from[String, Build_Job.Result], + Map.from[String, Result], { res => val name = res.string(Results.name) + val worker_uuid = res.string(Results.worker_uuid) + val build_uuid = res.string(Results.build_uuid) val hostname = res.string(Results.hostname) val numa_node = res.get_int(Results.numa_node) + val node_info = Host.Node_Info(hostname, numa_node) + val rc = res.int(Results.rc) val out = res.string(Results.out) val err = res.string(Results.err) val timing = res.timing( Results.timing_elapsed, Results.timing_cpu, Results.timing_gc) - val node_info = Host.Node_Info(hostname, numa_node) val process_result = Process_Result(rc, out_lines = split_lines(out), err_lines = split_lines(err), timing = timing) - name -> Build_Job.Result(node_info, process_result) + + val output_shasum = SHA1.fake_shasum(res.string(Results.output_shasum)) + val current = res.bool(Results.current) + + name -> + Result(name, worker_uuid, build_uuid, node_info, process_result, output_shasum, current) } ) def update_results(db: SQL.Database, results: State.Results): Boolean = { val old_results = read_results_domain(db) - val insert = results.iterator.filterNot(p => old_results.contains(p._1)).toList + val insert = results.valuesIterator.filterNot(res => old_results.contains(res.name)).toList - for ((name, result) <- insert) { - val node_info = result.node_info + for (result <- insert) { val process_result = result.process_result db.execute_statement(Results.table.insert(), body = { stmt => - stmt.string(1) = name - stmt.string(2) = node_info.hostname - stmt.int(3) = node_info.numa_node - stmt.int(4) = process_result.rc - stmt.string(5) = cat_lines(process_result.out_lines) - stmt.string(6) = cat_lines(process_result.err_lines) - stmt.long(7) = process_result.timing.elapsed.ms - stmt.long(8) = process_result.timing.cpu.ms - stmt.long(9) = process_result.timing.gc.ms + stmt.string(1) = result.name + stmt.string(2) = result.worker_uuid + stmt.string(3) = result.build_uuid + stmt.string(4) = result.node_info.hostname + stmt.int(5) = result.node_info.numa_node + stmt.int(6) = process_result.rc + stmt.string(7) = cat_lines(process_result.out_lines) + stmt.string(8) = cat_lines(process_result.err_lines) + stmt.long(9) = process_result.timing.elapsed.ms + stmt.long(10) = process_result.timing.cpu.ms + stmt.long(11) = process_result.timing.gc.ms + stmt.string(12) = result.output_shasum.toString + stmt.bool(13) = result.current }) } insert.nonEmpty } /* collective operations */ val all_tables: SQL.Tables = SQL.Tables( Base.table, Workers.table, Progress.table, Sessions.table, Pending.table, Running.table, Results.table, Host.Data.Node_Info.table) def update_database( db: SQL.Database, worker_uuid: String, build_uuid: String, hostname: String, state: State ): State = { val changed = List( update_sessions(db, state.sessions), update_pending(db, state.pending), update_running(db, state.running), update_results(db, state.results), Host.Data.update_numa_next(db, hostname, state.numa_next)) val serial0 = serial_max(db) val serial = if (changed.exists(identity)) State.inc_serial(serial0) else serial0 stamp_worker(db, worker_uuid, serial) state.set_serial(serial).set_workers(read_workers(db)) } } } /** main process **/ class Build_Process( protected final val build_context: Build_Process.Context, protected final val build_progress: Progress ) extends AutoCloseable { /* context */ protected final val store: Sessions.Store = build_context.store protected final val build_options: Options = store.options protected final val build_deps: Sessions.Deps = build_context.build_deps protected final val build_uuid: String = build_context.build_uuid protected final val worker_uuid: String = UUID.random().toString override def toString: String = "Build_Process(worker_uuid = " + quote(worker_uuid) + ", build_uuid = " + quote(build_uuid) + if_proper(build_context.master, ", master = true") + ")" /* global state: internal var vs. external database */ private var _state: Build_Process.State = init_state(Build_Process.State()) private val _database: Option[SQL.Database] = store.open_build_database() def close(): Unit = synchronized { _database.foreach(_.close()) } protected def synchronized_database[A](body: => A): A = synchronized { _database match { case None => body case Some(db) => @tailrec def loop(): A = { val sync_progress = db.transaction_lock(Build_Process.Data.all_tables) { val (messages, sync) = Build_Process.Data.sync_progress( db, _state.progress_seen, build_uuid, build_progress) if (sync) Left(body) else Right(messages) } sync_progress match { case Left(res) => res case Right(messages) => for ((message_serial, message) <- messages) { _state = _state.progress_serial(message_serial = message_serial) if (build_progress.do_output(message)) build_progress.output(message) } loop() } } loop() } } private def sync_database(): Unit = synchronized_database { for (db <- _database) { _state = Build_Process.Data.update_database( db, worker_uuid, build_uuid, build_context.hostname, _state) } } /* progress backed by database */ private def progress_output(message: Progress.Message, build_progress_output: => Unit): Unit = { synchronized_database { _state = _state.inc_serial.progress_serial() for (db <- _database) { Build_Process.Data.write_progress(db, _state.serial, message, build_uuid) Build_Process.Data.stamp_worker(db, worker_uuid, _state.serial) } build_progress_output } } protected object progress extends Progress { override def verbose: Boolean = build_progress.verbose override def output(message: Progress.Message): Unit = progress_output(message, if (do_output(message)) build_progress.output(message)) override def theory(theory: Progress.Theory): Unit = progress_output(theory.message, build_progress.theory(theory)) override def nodes_status(nodes_status: Document_Status.Nodes_Status): Unit = build_progress.nodes_status(nodes_status) override def stop(): Unit = build_progress.stop() override def stopped: Boolean = build_progress.stopped } protected val log: Logger = Logger.make_system_log(progress, build_options) /* policy operations */ protected def init_state(state: Build_Process.State): Build_Process.State = { val sessions1 = build_context.sessions.foldLeft(state.sessions) { case (map, (name, session)) => if (state.sessions.isDefinedAt(name)) map else map + (name -> session) } val old_pending = state.pending.iterator.map(_.name).toSet val new_pending = List.from( for { (name, session_context) <- build_context.sessions.iterator if !old_pending(name) } yield Build_Process.Task(name, session_context.deps)) val pending1 = new_pending ::: state.pending state.copy(sessions = sessions1, pending = pending1) } protected def next_job(state: Build_Process.State): Option[String] = if (progress.stopped || state.running.size < build_context.max_jobs) { state.pending.filter(entry => entry.is_ready && !state.is_running(entry.name)) .sortBy(_.name)(build_context.ordering) .headOption.map(_.name) } else None protected def start_session(state: Build_Process.State, session_name: String): Build_Process.State = { val ancestor_results = for (a <- build_context.sessions(session_name).ancestors) yield state.results(a) val input_shasum = if (ancestor_results.isEmpty) ML_Process.bootstrap_shasum() else SHA1.flat_shasum(ancestor_results.map(_.output_shasum)) val store_heap = build_context.store_heap(session_name) val (current, output_shasum) = store.check_output(session_name, sources_shasum = build_context.sources_shasum(session_name), input_shasum = input_shasum, fresh_build = build_context.fresh_build, store_heap = store_heap) val all_current = current && ancestor_results.forall(_.current) + val result_name = (session_name, worker_uuid, build_uuid) + if (all_current) { state .remove_pending(session_name) - .make_result(session_name, Process_Result.ok, output_shasum, current = true) + .make_result(result_name, Process_Result.ok, output_shasum, current = true) } else if (build_context.no_build) { progress.echo("Skipping " + session_name + " ...", verbose = true) state. remove_pending(session_name). - make_result(session_name, Process_Result.error, output_shasum) + make_result(result_name, Process_Result.error, output_shasum) } else if (progress.stopped || !ancestor_results.forall(_.ok)) { progress.echo(session_name + " CANCELLED") state .remove_pending(session_name) - .make_result(session_name, Process_Result.undefined, output_shasum) + .make_result(result_name, Process_Result.undefined, output_shasum) } else { val (numa_node, state1) = state.next_numa_node(build_context.numa_nodes) val node_info = Host.Node_Info(build_context.hostname, numa_node) progress.echo( (if (store_heap) "Building " else "Running ") + session_name + if_proper(node_info.numa_node, " on " + node_info) + " ...") store.init_output(session_name) val build = Build_Job.start_session(build_context, progress, log, build_deps.background(session_name), input_shasum, node_info) val job = Build_Process.Job(session_name, worker_uuid, build_uuid, node_info, Some(build)) state1.add_running(job) } } /* build process roles */ final def is_session_name(job_name: String): Boolean = !Long_Name.is_qualified(job_name) final def start_build(): Unit = synchronized_database { for (db <- _database) { Build_Process.Data.start_build(db, build_uuid, build_context.ml_platform, build_context.sessions_structure.session_prefs, progress.stopped) } } final def stop_build(): Unit = synchronized_database { for (db <- _database) { Build_Process.Data.stop_build(db, build_uuid) } } final def start_worker(): Unit = synchronized_database { for (db <- _database) { val java = ProcessHandle.current() val java_pid = java.pid val java_start = Date.instant(java.info.startInstant.get) val serial = Build_Process.Data.start_worker( db, worker_uuid, build_uuid, build_context.hostname, java_pid, java_start) _state = _state.set_serial(serial) } } final def stop_worker(): Unit = synchronized_database { for (db <- _database) { Build_Process.Data.stamp_worker(db, worker_uuid, _state.serial, stop = true) } } /* run */ def run(): Map[String, Process_Result] = { def finished(): Boolean = synchronized_database { _state.finished } def sleep(): Unit = Isabelle_Thread.interrupt_handler(_ => progress.stop()) { build_options.seconds("editor_input_delay").sleep() } def start_job(): Boolean = synchronized_database { next_job(_state) match { case Some(name) => if (is_session_name(name)) { _state = start_session(_state, name) true } else error("Unsupported build job name " + quote(name)) case None => false } } if (finished()) { progress.echo_warning("Nothing to build") Map.empty[String, Process_Result] } else { if (build_context.master) start_build() start_worker() if (build_context.master && !build_context.worker_active) { progress.echo("Waiting for external workers ...") } try { while (!finished()) { synchronized_database { if (progress.stopped) _state.stop_running() for (job <- _state.finished_running()) { + val result_name = (job.name, worker_uuid, build_uuid) val (process_result, output_shasum) = job.build.get.join _state = _state. remove_pending(job.name). remove_running(job.name). - make_result(job.name, process_result, output_shasum, node_info = job.node_info) + make_result(result_name, process_result, output_shasum, node_info = job.node_info) } } if (!start_job()) { sync_database() sleep() } } } finally { stop_worker() if (build_context.master) stop_build() } synchronized_database { for ((name, result) <- _state.results) yield name -> result.process_result } } } }