src/diagnostics/archivist/src/archivist.rs

// Copyright 2020 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

use {
    crate::{
        accessor::ArchiveAccessor,
        archive, configs, constants,
        constants::INSPECT_LOG_WINDOW_SIZE,
        container::ComponentIdentity,
        diagnostics,
        events::{
            source_registry::EventSourceRegistry,
            sources::{StaticEventStream, UnattributedLogSinkSource},
            types::{
                ComponentEvent, ComponentEventStream, DiagnosticsReadyEvent, EventMetadata,
                EventSource,
            },
        },
        logs::{budget::BudgetManager, redact::Redactor, socket::LogMessageSocket},
        pipeline::Pipeline,
        repository::DataRepo,
    },
    anyhow::Error,
    fidl::{endpoints::RequestStream, AsyncChannel},
    fidl_fuchsia_diagnostics::Selector,
    fidl_fuchsia_diagnostics_test::{ControllerRequest, ControllerRequestStream},
    fidl_fuchsia_process_lifecycle::{LifecycleRequest, LifecycleRequestStream},
    fidl_fuchsia_sys2::EventSourceMarker,
    fuchsia_async::{self as fasync, Task},
    fuchsia_component::{
        client::connect_to_service,
        server::{ServiceFs, ServiceObj, ServiceObjTrait},
    },
    fuchsia_inspect::{component, health::Reporter},
    fuchsia_inspect_contrib::{inspect_log, nodes::BoundedListNode},
    fuchsia_runtime::{take_startup_handle, HandleInfo, HandleType},
    fuchsia_zircon as zx,
    futures::{
        channel::mpsc,
        future::{self, abortable},
        prelude::*,
    },
    io_util,
    parking_lot::{Mutex, RwLock},
    std::{
        path::{Path, PathBuf},
        sync::{Arc, Weak},
    },
    tracing::{debug, error, info, warn},
};

/// Options for ingesting logs.
pub struct LogOpts {
    /// Whether or not logs coming from v2 components should be ingested.
    pub ingest_v2_logs: bool,
}

/// Responsible for initializing an `Archivist` instance. Supports multiple configurations by
/// either calling or not calling methods on the builder like `install_controller_service`.
pub struct ArchivistBuilder {
    /// Archive state, including the diagnostics repo which currently stores all logs.
    state: ArchivistState,

    /// True if pipeline exists.
    pipeline_exists: bool,

    /// Store for safe keeping,
    _pipeline_nodes: Vec<fuchsia_inspect::Node>,

    // Store for safe keeping.
    _pipeline_configs: Vec<configs::PipelineConfig>,

    /// ServiceFs object to server outgoing directory.
    fs: ServiceFs<ServiceObj<'static, ()>>,

    /// Receiver for stream which will process LogSink connections.
    log_receiver: mpsc::UnboundedReceiver<Task<()>>,

    /// Sender which is used to close the stream of LogSink connections.
    ///
    /// Clones of the sender keep the receiver end of the channel open. As soon
    /// as all clones are dropped or disconnected, the receiver will close. The
    /// receiver must close for `Archivist::run` to return gracefully.
    log_sender: mpsc::UnboundedSender<Task<()>>,

    /// Receiver for stream which will process Log connections.
    listen_receiver: mpsc::UnboundedReceiver<Task<()>>,

    /// Sender which is used to close the stream of Log connections after log_sender
    /// completes.
    ///
    /// Clones of the sender keep the receiver end of the channel open. As soon
    /// as all clones are dropped or disconnected, the receiver will close. The
    /// receiver must close for `Archivist::run` to return gracefully.
    listen_sender: mpsc::UnboundedSender<Task<()>>,

    /// Listes for events coming from v1 and v2.
    event_source_registry: EventSourceRegistry,

    /// Recieve stop signal to kill this archivist.
    stop_recv: Option<mpsc::Receiver<()>>,
}

impl ArchivistBuilder {
    /// Creates new instance, sets up inspect and adds 'archive' directory to output folder.
    /// Also installs `fuchsia.diagnostics.Archive` service.
    /// Call `install_log_services`, `add_event_source`.
    pub fn new(archivist_configuration: configs::Config) -> Result<Self, Error> {
        let (log_sender, log_receiver) = mpsc::unbounded();
        let (listen_sender, listen_receiver) = mpsc::unbounded();

        let mut fs = ServiceFs::new();
        diagnostics::serve(&mut fs)?;

        let writer = archivist_configuration.archive_path.as_ref().and_then(|archive_path| {
            maybe_create_archive(&mut fs, archive_path)
                .or_else(|e| {
                    // TODO(fxbug.dev/57271): this is not expected in regular builds of the archivist. It's
                    // happening when starting the zircon_guest (fx shell guest launch zircon_guest)
                    // We'd normally fail if we couldn't create the archive, but for now we include
                    // a warning.
                    warn!(
                        path = %archive_path.display(), ?e,
                        "Failed to create archive"
                    );
                    Err(e)
                })
                .ok()
        });

        let pipelines_node = diagnostics::root().create_child("pipelines");
        let feedback_pipeline_node = pipelines_node.create_child("feedback");
        let legacy_pipeline_node = pipelines_node.create_child("legacy_metrics");
        let mut feedback_config = configs::PipelineConfig::from_directory(
            "/config/data/feedback",
            configs::EmptyBehavior::DoNotFilter,
        );
        feedback_config.record_to_inspect(&feedback_pipeline_node);
        let mut legacy_config = configs::PipelineConfig::from_directory(
            "/config/data/legacy_metrics",
            configs::EmptyBehavior::Disable,
        );
        legacy_config.record_to_inspect(&legacy_pipeline_node);
        // Do not set the state to error if the pipelines simply do not exist.
        let pipeline_exists = !((Path::new("/config/data/feedback").is_dir()
            && feedback_config.has_error())
            || (Path::new("/config/data/legacy_metrics").is_dir() && legacy_config.has_error()));

        let logs_budget =
            BudgetManager::new(archivist_configuration.logs.max_cached_original_bytes);
        let diagnostics_repo = DataRepo::new(&logs_budget, diagnostics::root());

        // The Inspect Repository offered to the ALL_ACCESS pipeline. This
        // repository is unique in that it has no statically configured
        // selectors, meaning all diagnostics data is visible.
        // This should not be used for production services.
        // TODO(fxbug.dev/55735): Lock down this protocol using allowlists.
        let all_access_pipeline =
            Arc::new(RwLock::new(Pipeline::new(None, Redactor::noop(), diagnostics_repo.clone())));

        // The Inspect Repository offered to the Feedback pipeline. This repository applies
        // static selectors configured under config/data/feedback to inspect exfiltration.
        let (feedback_static_selectors, feedback_redactor) = if !feedback_config.disable_filtering {
            (
                feedback_config.take_inspect_selectors().map(|selectors| {
                    selectors
                        .into_iter()
                        .map(|selector| Arc::new(selector))
                        .collect::<Vec<Arc<Selector>>>()
                }),
                Redactor::with_static_patterns(),
            )
        } else {
            (None, Redactor::noop())
        };

        let feedback_pipeline = Arc::new(RwLock::new(Pipeline::new(
            feedback_static_selectors,
            feedback_redactor,
            diagnostics_repo.clone(),
        )));

        // The Inspect Repository offered to the LegacyMetrics
        // pipeline. This repository applies static selectors configured
        // under config/data/legacy_metrics to inspect exfiltration.
        let legacy_metrics_pipeline = Arc::new(RwLock::new(Pipeline::new(
            match legacy_config.disable_filtering {
                false => legacy_config.take_inspect_selectors().map(|selectors| {
                    selectors
                        .into_iter()
                        .map(|selector| Arc::new(selector))
                        .collect::<Vec<Arc<Selector>>>()
                }),
                true => None,
            },
            Redactor::noop(),
            diagnostics_repo.clone(),
        )));

        // TODO(fxbug.dev/55736): Refactor this code so that we don't store
        // diagnostics data N times if we have N pipelines. We should be
        // storing a single copy regardless of the number of pipelines.
        let archivist_state = ArchivistState::new(
            archivist_configuration,
            vec![
                all_access_pipeline.clone(),
                feedback_pipeline.clone(),
                legacy_metrics_pipeline.clone(),
            ],
            diagnostics_repo,
            logs_budget,
            writer,
        )?;

        let all_accessor_stats = Arc::new(diagnostics::AccessorStats::new(
            component::inspector().root().create_child("all_archive_accessor"),
        ));

        let feedback_accessor_stats = Arc::new(diagnostics::AccessorStats::new(
            component::inspector().root().create_child("feedback_archive_accessor"),
        ));

        let legacy_accessor_stats = Arc::new(diagnostics::AccessorStats::new(
            component::inspector().root().create_child("legacy_metrics_archive_accessor"),
        ));

        let sender_for_accessor = listen_sender.clone();
        let sender_for_feedback = listen_sender.clone();
        let sender_for_legacy = listen_sender.clone();
        fs.dir("svc")
            .add_fidl_service(move |stream| {
                debug!("fuchsia.diagnostics.ArchiveAccessor connection");
                let all_archive_accessor =
                    ArchiveAccessor::new(all_access_pipeline.clone(), all_accessor_stats.clone());
                all_archive_accessor
                    .spawn_archive_accessor_server(stream, sender_for_accessor.clone());
            })
            .add_fidl_service_at(constants::FEEDBACK_ARCHIVE_ACCESSOR_NAME, move |chan| {
                debug!("fuchsia.diagnostics.FeedbackArchiveAccessor connection");
                let feedback_archive_accessor = ArchiveAccessor::new(
                    feedback_pipeline.clone(),
                    feedback_accessor_stats.clone(),
                );
                feedback_archive_accessor
                    .spawn_archive_accessor_server(chan, sender_for_feedback.clone());
            })
            .add_fidl_service_at(constants::LEGACY_METRICS_ARCHIVE_ACCESSOR_NAME, move |chan| {
                debug!("fuchsia.diagnostics.LegacyMetricsAccessor connection");
                let legacy_archive_accessor = ArchiveAccessor::new(
                    legacy_metrics_pipeline.clone(),
                    legacy_accessor_stats.clone(),
                );
                legacy_archive_accessor
                    .spawn_archive_accessor_server(chan, sender_for_legacy.clone());
            });

        let events_node = diagnostics::root().create_child("event_stats");
        Ok(Self {
            fs,
            state: archivist_state,
            log_receiver,
            log_sender,
            listen_receiver,
            listen_sender,
            pipeline_exists,
            _pipeline_nodes: vec![pipelines_node, feedback_pipeline_node, legacy_pipeline_node],
            _pipeline_configs: vec![feedback_config, legacy_config],
            event_source_registry: EventSourceRegistry::new(events_node),
            stop_recv: None,
        })
    }

    pub fn data_repo(&self) -> &DataRepo {
        &self.state.diagnostics_repo
    }

    pub fn log_sender(&self) -> &mpsc::UnboundedSender<Task<()>> {
        &self.log_sender
    }

    // TODO(fxbug.dev/72046) delete when netemul no longer using
    pub fn consume_own_logs(&self, socket: zx::Socket) {
        let container = self.data_repo().write().get_own_log_container();
        fasync::Task::spawn(async move {
            let log_stream =
                LogMessageSocket::new(socket, container.identity.clone(), container.stats.clone())
                    .expect("failed to create internal LogMessageSocket");
            container.drain_messages(log_stream).await;
            unreachable!();
        })
        .detach();
    }

    /// Install controller service.
    pub fn install_controller_service(&mut self) -> &mut Self {
        let (stop_sender, stop_recv) = mpsc::channel(0);
        self.fs
            .dir("svc")
            .add_fidl_service(move |stream| Self::spawn_controller(stream, stop_sender.clone()));
        self.stop_recv = Some(stop_recv);
        debug!("Controller services initialized.");
        self
    }

    /// The spawned controller listens for the stop request and forwards that to the archivist stop
    /// channel if received.
    fn spawn_controller(mut stream: ControllerRequestStream, mut stop_sender: mpsc::Sender<()>) {
        fasync::Task::spawn(
            async move {
                while let Some(ControllerRequest::Stop { .. }) = stream.try_next().await? {
                    debug!("Stop request received.");
                    stop_sender.send(()).await.ok();
                    break;
                }
                Ok(())
            }
            .map(|o: Result<(), fidl::Error>| {
                if let Err(e) = o {
                    error!(%e, "error serving controller");
                }
            }),
        )
        .detach();
    }

    fn take_lifecycle_channel() -> LifecycleRequestStream {
        let lifecycle_handle_info = HandleInfo::new(HandleType::Lifecycle, 0);
        let lifecycle_handle = take_startup_handle(lifecycle_handle_info)
            .expect("must have been provided a lifecycle channel in procargs");
        let x: zx::Channel = lifecycle_handle.into();
        let async_x = AsyncChannel::from(
            fasync::Channel::from_channel(x).expect("Async channel conversion failed."),
        );
        LifecycleRequestStream::from_channel(async_x)
    }

    pub fn install_lifecycle_listener(&mut self) -> &mut Self {
        let (mut stop_sender, stop_recv) = mpsc::channel(0);
        let mut req_stream = Self::take_lifecycle_channel();

        Task::spawn(async move {
            debug!("Awaiting request to close");
            while let Some(LifecycleRequest::Stop { .. }) =
                req_stream.try_next().await.expect("Failure receiving lifecycle FIDL message")
            {
                info!("Initiating shutdown.");
                stop_sender.send(()).await.unwrap();
            }
        })
        .detach();

        self.stop_recv = Some(stop_recv);
        debug!("Lifecycle listener initialized.");
        self
    }

    /// Installs `LogSink` and `Log` services. Panics if called twice.
    /// # Arguments:
    /// * `log_connector` - If provided, install log connector.
    pub async fn install_log_services(&mut self, opts: LogOpts) -> &mut Self {
        let data_repo_1 = self.data_repo().clone();
        let listen_sender = self.listen_sender.clone();

        let unattributed_log_sink_source = UnattributedLogSinkSource::new();
        let unattributed_log_sink_publisher = unattributed_log_sink_source.get_publisher();
        self.event_source_registry
            .add_source("unattributed_log_sink", Box::new(unattributed_log_sink_source))
            .await;

        self.fs
            .dir("svc")
            .add_fidl_service(move |stream| {
                debug!("fuchsia.logger.Log connection");
                data_repo_1.clone().handle_log(stream, listen_sender.clone());
            })
            .add_fidl_service(move |stream| {
                debug!("unattributed fuchsia.logger.LogSink connection");
                let mut publisher = unattributed_log_sink_publisher.clone();
                // TODO(fxbug.dev/67769): get rid of this Task spawn since it introduces a small
                // window in which we might lose LogSinks.
                fasync::Task::spawn(async move {
                    publisher.send(stream).await.unwrap_or_else(|err| {
                        error!(?err, "Failed to add unattributed LogSink connection")
                    })
                })
                .detach();
            });
        if opts.ingest_v2_logs {
            debug!("fuchsia.sys.EventStream connection");
            let event_source = connect_to_service::<EventSourceMarker>().unwrap();
            match event_source.take_static_event_stream("EventStream").await {
                Ok(Ok(event_stream)) => {
                    let event_stream = event_stream.into_stream().unwrap();
                    self.event_source_registry
                        .add_source(
                            "v2_static_event_stream",
                            Box::new(StaticEventStream::new(event_stream)),
                        )
                        .await;
                }
                Ok(Err(err)) => debug!(?err, "Failed to open event stream"),
                Err(err) => debug!(?err, "Failed to send request to take event stream"),
            }
        }
        debug!("Log services initialized.");
        self
    }

    // Sets event provider which is used to collect component events, Panics if called twice.
    pub async fn add_event_source(
        &mut self,
        name: impl Into<String>,
        source: Box<dyn EventSource>,
    ) -> &mut Self {
        let name = name.into();
        debug!("{} event source initialized", &name);
        self.event_source_registry.add_source(name, source).await;
        self
    }

    /// Run archivist to completion.
    /// # Arguments:
    /// * `outgoing_channel`- channel to serve outgoing directory on.
    pub async fn run(mut self, outgoing_channel: zx::Channel) -> Result<(), Error> {
        debug!("Running Archivist.");

        let data_repo = { self.data_repo().clone() };
        self.fs.serve_connection(outgoing_channel)?;
        // Start servcing all outgoing services.
        let run_outgoing = self.fs.collect::<()>();
        // collect events.
        let events = self.event_source_registry.take_stream().await.expect("Created event stream");
        let logs_budget = self.state.logs_budget.handle();
        let run_event_collection_task = fasync::Task::spawn(Self::collect_component_events(
            events,
            self.log_sender.clone(),
            self.state,
            self.pipeline_exists,
        ));

        // Process messages from log sink.
        let log_receiver = self.log_receiver;
        let listen_receiver = self.listen_receiver;
        let all_msg = async {
            log_receiver.for_each_concurrent(None, |rx| async move { rx.await }).await;
            debug!("Log ingestion stopped.");
            data_repo.terminate_logs();
            logs_budget.terminate();
            debug!("Flushing to listeners.");
            listen_receiver.for_each_concurrent(None, |rx| async move { rx.await }).await;
            debug!("Log listeners and batch iterators stopped.");
        };

        let (abortable_fut, abort_handle) = abortable(run_outgoing);

        let mut listen_sender = self.listen_sender;
        let mut log_sender = self.log_sender;
        let event_source_registry = self.event_source_registry;
        let stop_fut = match self.stop_recv {
            Some(stop_recv) => async move {
                stop_recv.into_future().await;
                event_source_registry.terminate();
                run_event_collection_task.await;
                listen_sender.disconnect();
                log_sender.disconnect();
                abort_handle.abort()
            }
            .left_future(),
            None => future::ready(()).right_future(),
        };

        debug!("Entering core loop.");
        // Combine all three futures into a main future.
        future::join3(abortable_fut, stop_fut, all_msg).map(|_| Ok(())).await
    }

    async fn collect_component_events(
        events: ComponentEventStream,
        log_sender: mpsc::UnboundedSender<Task<()>>,
        state: ArchivistState,
        pipeline_exists: bool,
    ) {
        if !pipeline_exists {
            component::health().set_unhealthy("Pipeline config has an error");
        } else {
            component::health().set_ok();
        }
        state.run(events, log_sender).await
    }
}

/// ArchivistState owns the tools needed to persist data
/// to the archive, as well as the service-specific repositories
/// that are populated by the archivist server and exposed in the
/// service sessions.
pub struct ArchivistState {
    /// Writer for the archive. If a path was not configured it will be `None`.
    writer: Option<archive::ArchiveWriter>,
    log_node: BoundedListNode,
    configuration: configs::Config,
    diagnostics_pipelines: Vec<Arc<RwLock<Pipeline>>>,
    pub diagnostics_repo: DataRepo,

    /// The overall capacity we enforce for log messages across containers.
    logs_budget: BudgetManager,
}

impl ArchivistState {
    pub fn new(
        configuration: configs::Config,
        diagnostics_pipelines: Vec<Arc<RwLock<Pipeline>>>,
        diagnostics_repo: DataRepo,
        logs_budget: BudgetManager,
        writer: Option<archive::ArchiveWriter>,
    ) -> Result<Self, Error> {
        let mut log_node = BoundedListNode::new(
            diagnostics::root().create_child("events"),
            INSPECT_LOG_WINDOW_SIZE,
        );

        inspect_log!(log_node, event: "Archivist started");

        Ok(ArchivistState {
            writer,
            log_node,
            configuration,
            diagnostics_pipelines,
            diagnostics_repo,
            logs_budget,
        })
    }

    pub async fn run(
        self,
        mut events: ComponentEventStream,
        log_sender: mpsc::UnboundedSender<Task<()>>,
    ) {
        let state = Arc::new(Mutex::new(self));
        let remover = ComponentRemover { inner: Arc::downgrade(&state) };
        state.lock().logs_budget.set_remover(remover);

        let archivist = Archivist { state, log_sender };
        while let Some(event) = events.next().await {
            archivist.clone().process_event(event).await.unwrap_or_else(|e| {
            let mut state = archivist.lock();
            inspect_log!(state.log_node, event: "Failed to log event", result: format!("{:?}", e));
            error!(?e, "Failed to log event");
        });
        }
    }

    fn maybe_remove_component(&mut self, identity: &ComponentIdentity) {
        if !self.diagnostics_repo.is_live(identity) {
            debug!(%identity, "Removing component from repository.");
            self.diagnostics_repo.write().data_directories.remove(&identity.unique_key);

            // TODO(fxbug.dev/55736): The pipeline specific updates should be happening asynchronously.
            for pipeline in &self.diagnostics_pipelines {
                pipeline.write().remove(&identity.relative_moniker);
            }
        }
    }
}

#[derive(Clone)]
pub struct Archivist {
    state: Arc<Mutex<ArchivistState>>,
    log_sender: mpsc::UnboundedSender<Task<()>>,
}

impl std::ops::Deref for Archivist {
    type Target = Mutex<ArchivistState>;
    fn deref(&self) -> &Self::Target {
        &*self.state
    }
}

impl Archivist {
    async fn populate_inspect_repo(&self, diagnostics_ready_data: DiagnosticsReadyEvent) {
        // The DiagnosticsReadyEvent should always contain a directory_proxy. Its existence
        // as an Option is only to support mock objects for equality in tests.
        let diagnostics_proxy = diagnostics_ready_data.directory.unwrap();

        let identity = diagnostics_ready_data.metadata.identity.clone();

        let locked_state = &self.lock();

        // Update the central repository to reference the new diagnostics source.
        locked_state
            .diagnostics_repo
            .write()
            .add_inspect_artifacts(
                identity.clone(),
                diagnostics_proxy,
                diagnostics_ready_data.metadata.timestamp.clone(),
            )
            .unwrap_or_else(|e| {
                warn!(%identity, ?e, "Failed to add inspect artifacts to repository");
            });

        // Let each pipeline know that a new component arrived, and allow the pipeline
        // to eagerly bucket static selectors based on that component's moniker.
        // TODO(fxbug.dev/55736): The pipeline specific updates should be happening asynchronously.
        for pipeline in &locked_state.diagnostics_pipelines {
            pipeline.write().add_inspect_artifacts(&identity.relative_moniker).unwrap_or_else(
                |e| {
                    warn!(%identity, ?e, "Failed to add inspect artifacts to pipeline wrapper");
                },
            );
        }
    }

    /// Updates the central repository to reference the new diagnostics source.
    fn add_new_component(
        &self,
        identity: ComponentIdentity,
        event_timestamp: zx::Time,
        component_start_time: Option<zx::Time>,
    ) {
        if let Err(e) = self.lock().diagnostics_repo.write().add_new_component(
            identity.clone(),
            event_timestamp,
            component_start_time,
        ) {
            error!(%identity, ?e, "Failed to add new component to repository");
        }
    }

    fn mark_component_stopped(&self, identity: &ComponentIdentity) {
        let mut this = self.state.lock();
        // TODO(fxbug.dev/53939): Get inspect data from repository before removing
        // for post-mortem inspection.
        this.diagnostics_repo.write().mark_stopped(&identity.unique_key);
        this.maybe_remove_component(identity);
    }

    async fn process_event(self, event: ComponentEvent) -> Result<(), Error> {
        match event {
            ComponentEvent::Start(start) => {
                let archived_metadata = start.metadata.clone();
                debug!(identity = %start.metadata.identity, "Adding new component.");
                self.add_new_component(start.metadata.identity, start.metadata.timestamp, None);
                self.archive_event("START", archived_metadata).await
            }
            ComponentEvent::Running(running) => {
                let archived_metadata = running.metadata.clone();
                debug!(identity = %running.metadata.identity, "Component is running.");
                self.add_new_component(
                    running.metadata.identity,
                    running.metadata.timestamp,
                    Some(running.component_start_time),
                );
                self.archive_event("RUNNING", archived_metadata.clone()).await
            }
            ComponentEvent::Stop(stop) => {
                debug!(identity = %stop.metadata.identity, "Component stopped");
                self.mark_component_stopped(&stop.metadata.identity);
                self.archive_event("STOP", stop.metadata).await
            }
            ComponentEvent::DiagnosticsReady(diagnostics_ready) => {
                debug!(
                    identity = %diagnostics_ready.metadata.identity,
                    "Diagnostics directory is ready.",
                );
                self.populate_inspect_repo(diagnostics_ready).await;
                Ok(())
            }
            ComponentEvent::LogSinkRequested(event) => {
                let locked_state = self.state.lock();
                let data_repo = &locked_state.diagnostics_repo;
                let container = data_repo.write().get_log_container(event.metadata.identity);
                container.handle_log_sink(event.requests, self.log_sender.clone());
                Ok(())
            }
        }
    }

    async fn archive_event(
        &self,
        _event_name: &str,
        _event_data: EventMetadata,
    ) -> Result<(), Error> {
        let mut state = self.lock();
        let ArchivistState { writer, log_node, configuration, .. } = &mut *state;

        let writer = if let Some(w) = writer.as_mut() {
            w
        } else {
            return Ok(());
        };

        let max_archive_size_bytes = configuration.max_archive_size_bytes;
        let max_event_group_size_bytes = configuration.max_event_group_size_bytes;

        // TODO(fxbug.dev/53939): Get inspect data from repository before removing
        // for post-mortem inspection.
        //let log = writer.get_log().new_event(event_name, event_data.component_id);
        // if let Some(data_map) = event_data.component_data_map {
        //     for (path, object) in data_map {
        //         match object {
        //             InspectData::Empty
        //             | InspectData::DeprecatedFidl(_)
        //             | InspectData::Tree(_, None) => {}
        //             InspectData::Vmo(vmo) | InspectData::Tree(_, Some(vmo)) => {
        //                 let mut contents = vec![0u8; vmo.get_size()? as usize];
        //                 vmo.read(&mut contents[..], 0)?;

        //                 // Truncate the bytes down to the last non-zero 4096-byte page of data.
        //                 // TODO(fxbug.dev/4703): Handle truncation of VMOs without reading the whole thing.
        //                 let mut last_nonzero = 0;
        //                 for (i, v) in contents.iter().enumerate() {
        //                     if *v != 0 {
        //                         last_nonzero = i;
        //                     }
        //                 }
        //                 if last_nonzero % 4096 != 0 {
        //                     last_nonzero = last_nonzero + 4096 - last_nonzero % 4096;
        //                 }
        //                 contents.resize(last_nonzero, 0);

        //                 log = log.add_event_file(path, &contents);
        //             }
        //             InspectData::File(contents) => {
        //                 log = log.add_event_file(path, &contents);
        //             }
        //         }
        //     }
        // }

        let current_group_stats = writer.get_log().get_stats();

        if current_group_stats.size >= max_event_group_size_bytes {
            let (path, stats) = writer.rotate_log()?;
            inspect_log!(log_node, event:"Rotated log",
                     new_path: path.to_string_lossy().to_string());
            writer.add_group_stat(&path, stats);
        }

        let archived_size = writer.archived_size();
        let mut current_archive_size = current_group_stats.size + archived_size;
        if current_archive_size > max_archive_size_bytes {
            let dates = writer.get_archive().get_dates().unwrap_or_else(|e| {
                warn!("Garbage collection failure");
                inspect_log!(log_node, event: "Failed to get dates for garbage collection",
                         reason: format!("{:?}", e));
                vec![]
            });

            for date in dates {
                let groups =
                    writer.get_archive().get_event_file_groups(&date).unwrap_or_else(|e| {
                        warn!("Garbage collection failure");
                        inspect_log!(log_node, event: "Failed to get event file",
                             date: &date,
                             reason: format!("{:?}", e));
                        vec![]
                    });

                for group in groups {
                    let path = group.log_file_path();
                    match group.delete() {
                        Err(e) => {
                            inspect_log!(log_node, event: "Failed to remove group",
                                 path: &path,
                                 reason: format!(
                                     "{:?}", e));
                            continue;
                        }
                        Ok(stat) => {
                            current_archive_size -= stat.size;
                            writer.remove_group_stat(&PathBuf::from(&path));
                            inspect_log!(log_node, event: "Garbage collected group",
                                     path: &path,
                                     removed_files: stat.file_count as u64,
                                     removed_bytes: stat.size as u64);
                        }
                    };

                    if current_archive_size < max_archive_size_bytes {
                        return Ok(());
                    }
                }
            }
        }

        Ok(())
    }
}

/// Provides weak access to ArchivistState for BudgetManager to use when it pops the last message
/// from the buffer of a stopped component.
#[derive(Clone, Default)]
pub struct ComponentRemover {
    inner: Weak<Mutex<ArchivistState>>,
}

impl ComponentRemover {
    pub fn maybe_remove_component(&self, identity: &ComponentIdentity) {
        if let Some(this) = self.inner.upgrade() {
            this.lock().maybe_remove_component(identity);
        }
    }
}

fn maybe_create_archive<ServiceObjTy: ServiceObjTrait>(
    fs: &mut ServiceFs<ServiceObjTy>,
    archive_path: &PathBuf,
) -> Result<archive::ArchiveWriter, Error> {
    let writer = archive::ArchiveWriter::open(archive_path.clone())?;
    fs.add_remote(
        "archive",
        io_util::open_directory_in_namespace(
            &archive_path.to_string_lossy(),
            io_util::OPEN_RIGHT_READABLE | io_util::OPEN_RIGHT_WRITABLE,
        )?,
    );
    Ok(writer)
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::{constants::LEGACY_DEFAULT_MAXIMUM_CACHED_LOGS_BYTES, logs::testing::*};
    use fidl::endpoints::create_proxy;
    use fidl_fuchsia_diagnostics_test::ControllerMarker;
    use fidl_fuchsia_io as fio;
    use fio::DirectoryProxy;
    use fuchsia_async as fasync;
    use fuchsia_component::client::connect_to_protocol_at_dir_svc;
    use futures::channel::oneshot;

    fn init_archivist() -> ArchivistBuilder {
        let config = configs::Config {
            archive_path: None,
            max_archive_size_bytes: 10,
            max_event_group_size_bytes: 10,
            num_threads: 1,
            logs: configs::LogsConfig {
                max_cached_original_bytes: LEGACY_DEFAULT_MAXIMUM_CACHED_LOGS_BYTES,
            },
        };

        ArchivistBuilder::new(config).unwrap()
    }

    // run archivist and send signal when it dies.
    async fn run_archivist_and_signal_on_exit() -> (DirectoryProxy, oneshot::Receiver<()>) {
        let (directory, server_end) = create_proxy::<fio::DirectoryMarker>().unwrap();
        let mut archivist = init_archivist();
        archivist
            .install_log_services(LogOpts { ingest_v2_logs: false })
            .await
            .install_controller_service();
        let (signal_send, signal_recv) = oneshot::channel();
        fasync::Task::spawn(async move {
            archivist.run(server_end.into_channel()).await.expect("Cannot run archivist");
            signal_send.send(()).unwrap();
        })
        .detach();
        (directory, signal_recv)
    }

    // runs archivist and returns its directory.
    async fn run_archivist() -> DirectoryProxy {
        let (directory, server_end) = create_proxy::<fio::DirectoryMarker>().unwrap();
        let mut archivist = init_archivist();
        archivist.install_log_services(LogOpts { ingest_v2_logs: false }).await;
        fasync::Task::spawn(async move {
            archivist.run(server_end.into_channel()).await.expect("Cannot run archivist");
        })
        .detach();
        directory
    }

    #[fuchsia::test]
    async fn can_log_and_retrive_log() {
        let directory = run_archivist().await;
        let mut recv_logs = start_listener(&directory);

        let mut log_helper = LogSinkHelper::new(&directory);
        log_helper.write_log("my msg1");
        log_helper.write_log("my msg2");

        assert_eq!(
            vec! {Some("my msg1".to_owned()),Some("my msg2".to_owned())},
            vec! {recv_logs.next().await,recv_logs.next().await}
        );

        // new client can log
        let mut log_helper2 = LogSinkHelper::new(&directory);
        log_helper2.write_log("my msg1");
        log_helper.write_log("my msg2");

        let mut expected = vec!["my msg1".to_owned(), "my msg2".to_owned()];
        expected.sort();

        let mut actual = vec![recv_logs.next().await.unwrap(), recv_logs.next().await.unwrap()];
        actual.sort();

        assert_eq!(expected, actual);

        // can log after killing log sink proxy
        log_helper.kill_log_sink();
        log_helper.write_log("my msg1");
        log_helper.write_log("my msg2");

        assert_eq!(
            expected,
            vec! {recv_logs.next().await.unwrap(),recv_logs.next().await.unwrap()}
        );

        // can log from new socket cnonnection
        log_helper2.add_new_connection();
        log_helper2.write_log("my msg1");
        log_helper2.write_log("my msg2");

        assert_eq!(
            expected,
            vec! {recv_logs.next().await.unwrap(),recv_logs.next().await.unwrap()}
        );
    }

    /// Makes sure that implementaion can handle multiple sockets from same
    /// log sink.
    #[fuchsia::test]
    async fn log_from_multiple_sock() {
        let directory = run_archivist().await;
        let mut recv_logs = start_listener(&directory);

        let log_helper = LogSinkHelper::new(&directory);
        let sock1 = log_helper.connect();
        let sock2 = log_helper.connect();
        let sock3 = log_helper.connect();

        LogSinkHelper::write_log_at(&sock1, "msg sock1-1");
        LogSinkHelper::write_log_at(&sock2, "msg sock2-1");
        LogSinkHelper::write_log_at(&sock1, "msg sock1-2");
        LogSinkHelper::write_log_at(&sock3, "msg sock3-1");
        LogSinkHelper::write_log_at(&sock2, "msg sock2-2");

        let mut expected = vec![
            "msg sock1-1".to_owned(),
            "msg sock1-2".to_owned(),
            "msg sock2-1".to_owned(),
            "msg sock2-2".to_owned(),
            "msg sock3-1".to_owned(),
        ];
        expected.sort();

        let mut actual = vec![
            recv_logs.next().await.unwrap(),
            recv_logs.next().await.unwrap(),
            recv_logs.next().await.unwrap(),
            recv_logs.next().await.unwrap(),
            recv_logs.next().await.unwrap(),
        ];
        actual.sort();

        assert_eq!(expected, actual);
    }

    /// Stop API works
    #[fuchsia::test]
    async fn stop_works() {
        let (directory, signal_recv) = run_archivist_and_signal_on_exit().await;
        let mut recv_logs = start_listener(&directory);

        {
            // make sure we can write logs
            let log_sink_helper = LogSinkHelper::new(&directory);
            let sock1 = log_sink_helper.connect();
            LogSinkHelper::write_log_at(&sock1, "msg sock1-1");
            log_sink_helper.write_log("msg sock1-2");
            let mut expected = vec!["msg sock1-1".to_owned(), "msg sock1-2".to_owned()];
            expected.sort();
            let mut actual = vec![recv_logs.next().await.unwrap(), recv_logs.next().await.unwrap()];
            actual.sort();
            assert_eq!(expected, actual);

            //  Start new connections and sockets
            let log_sink_helper1 = LogSinkHelper::new(&directory);
            let sock2 = log_sink_helper.connect();
            // Write logs before calling stop
            log_sink_helper1.write_log("msg 1");
            log_sink_helper1.write_log("msg 2");
            let log_sink_helper2 = LogSinkHelper::new(&directory);

            let controller = connect_to_protocol_at_dir_svc::<ControllerMarker>(&directory)
                .expect("cannot connect to log proxy");
            controller.stop().unwrap();

            // make more socket connections and write to them and old ones.
            let sock3 = log_sink_helper2.connect();
            log_sink_helper2.write_log("msg 3");
            log_sink_helper2.write_log("msg 4");

            LogSinkHelper::write_log_at(&sock3, "msg 5");
            LogSinkHelper::write_log_at(&sock2, "msg 6");
            log_sink_helper.write_log("msg 7");
            LogSinkHelper::write_log_at(&sock1, "msg 8");

            LogSinkHelper::write_log_at(&sock2, "msg 9");
        } // kills all sockets and log_sink connections
        let mut expected = vec![];
        let mut actual = vec![];
        for i in 1..=9 {
            expected.push(format!("msg {}", i));
            actual.push(recv_logs.next().await.unwrap());
        }
        expected.sort();
        actual.sort();

        // make sure archivist is dead.
        signal_recv.await.unwrap();

        assert_eq!(expected, actual);
    }
}