Critical dependency logging (#4988)

* add metrics layer

* add metrics

* simplify getting the target

* make clippy happy

* fix typos

* unify deps under workspace

* make import statement shorter, fix typos

* enable warn by default, mark flag as deprecated

* do not exit on error when initializing logging fails

* revert exit on error

* adjust bootnode logging

* add logging layer

* non blocking file writer

* non blocking file writer

* add tracing visitor

* use target as is by default

* make libp2p events register correctly

* adjust repilcated cli help

* refactor tracing layer

* linting

* filesize

* log gossipsub, dont filter by log level

* turn on debug logs by default, remove deprecation warning

* truncate file, add timestamp, add unit test

* suppress output (#5)

* use tracing appender

* cleanup

* Add a task to remove old log files and upgrade to warn level

* Add the time feature for tokio

* Udeps and fmt

---------

Co-authored-by: Diva M <divma@protonmail.com>
Co-authored-by: Divma <26765164+divagant-martian@users.noreply.github.com>
Co-authored-by: Age Manning <Age@AgeManning.com>

common/logging/src/lib.rs

@@ -7,15 +7,21 @@ use lighthouse_metrics::{
 use slog::Logger;
 use slog_term::Decorator;
 use std::io::{Result, Write};
+use std::path::PathBuf;
 use std::time::{Duration, Instant};
+use tracing_appender::non_blocking::NonBlocking;
+use tracing_logging_layer::LoggingLayer;
+use tracing_subscriber::{layer::SubscriberExt, util::SubscriberInitExt};
 
 pub const MAX_MESSAGE_WIDTH: usize = 40;
 
 pub mod async_record;
 mod sse_logging_components;
+mod tracing_logging_layer;
 mod tracing_metrics_layer;
 
 pub use sse_logging_components::SSELoggingComponents;
+pub use tracing_logging_layer::cleanup_logging_task;
 pub use tracing_metrics_layer::MetricsLayer;
 
 /// The minimum interval between log messages indicating that a queue is full.
@@ -217,6 +223,48 @@ impl TimeLatch {
     }
 }
 
+pub fn create_tracing_layer(base_tracing_log_path: PathBuf, turn_on_terminal_logs: bool) {
+    let filter_layer = match tracing_subscriber::EnvFilter::try_from_default_env()
+        .or_else(|_| tracing_subscriber::EnvFilter::try_new("warn"))
+    {
+        Ok(filter) => filter,
+        Err(e) => {
+            eprintln!("Failed to initialize dependency logging {e}");
+            return;
+        }
+    };
+
+    let libp2p_writer =
+        tracing_appender::rolling::daily(base_tracing_log_path.clone(), "libp2p.log");
+    let discv5_writer =
+        tracing_appender::rolling::daily(base_tracing_log_path.clone(), "discv5.log");
+
+    let (libp2p_non_blocking_writer, libp2p_guard) = NonBlocking::new(libp2p_writer);
+    let (discv5_non_blocking_writer, discv5_guard) = NonBlocking::new(discv5_writer);
+
+    let custom_layer = LoggingLayer {
+        libp2p_non_blocking_writer,
+        libp2p_guard,
+        discv5_non_blocking_writer,
+        discv5_guard,
+    };
+
+    if let Err(e) = tracing_subscriber::fmt()
+        .with_env_filter(filter_layer)
+        .with_writer(move || {
+            tracing_subscriber::fmt::writer::OptionalWriter::<std::io::Stdout>::from(
+                turn_on_terminal_logs.then(std::io::stdout),
+            )
+        })
+        .finish()
+        .with(MetricsLayer)
+        .with(custom_layer)
+        .try_init()
+    {
+        eprintln!("Failed to initialize dependency logging {e}");
+    }
+}
+
 /// Return a logger suitable for test usage.
 ///
 /// By default no logs will be printed, but they can be enabled via

common/logging/src/tracing_logging_layer.rs

@@ -0,0 +1,115 @@
+use chrono::{naive::Days, prelude::*};
+use slog::{debug, warn};
+use std::io::Write;
+use tracing::Subscriber;
+use tracing_appender::non_blocking::{NonBlocking, WorkerGuard};
+use tracing_subscriber::layer::Context;
+use tracing_subscriber::Layer;
+
+pub struct LoggingLayer {
+    pub libp2p_non_blocking_writer: NonBlocking,
+    pub libp2p_guard: WorkerGuard,
+    pub discv5_non_blocking_writer: NonBlocking,
+    pub discv5_guard: WorkerGuard,
+}
+
+impl<S> Layer<S> for LoggingLayer
+where
+    S: Subscriber,
+{
+    fn on_event(&self, event: &tracing::Event<'_>, _ctx: Context<S>) {
+        let meta = event.metadata();
+        let log_level = meta.level();
+        let timestamp = Local::now().format("%Y-%m-%d %H:%M:%S").to_string();
+
+        let target = match meta.target().split_once("::") {
+            Some((crate_name, _)) => crate_name,
+            None => "unknown",
+        };
+
+        let mut writer = match target {
+            "libp2p_gossipsub" => self.libp2p_non_blocking_writer.clone(),
+            "discv5" => self.discv5_non_blocking_writer.clone(),
+            _ => return,
+        };
+
+        let mut visitor = LogMessageExtractor {
+            message: String::default(),
+        };
+
+        event.record(&mut visitor);
+        let message = format!("{} {} {}\n", timestamp, log_level, visitor.message);
+
+        if let Err(e) = writer.write_all(message.as_bytes()) {
+            eprintln!("Failed to write log: {}", e);
+        }
+    }
+}
+
+struct LogMessageExtractor {
+    message: String,
+}
+
+impl tracing_core::field::Visit for LogMessageExtractor {
+    fn record_debug(&mut self, _: &tracing_core::Field, value: &dyn std::fmt::Debug) {
+        self.message = format!("{} {:?}", self.message, value);
+    }
+}
+
+/// Creates a long lived async task that routinely deletes old tracing log files
+pub async fn cleanup_logging_task(path: std::path::PathBuf, log: slog::Logger) {
+    loop {
+        // Delay for 1 day and then prune old logs
+        tokio::time::sleep(std::time::Duration::from_secs(60 * 60 * 24)).await;
+
+        let Some(yesterday_date) = chrono::prelude::Local::now()
+            .naive_local()
+            .checked_sub_days(Days::new(1))
+        else {
+            warn!(log, "Could not calculate the current date");
+            return;
+        };
+
+        // Search for old log files
+        let dir = path.as_path();
+
+        if dir.is_dir() {
+            let Ok(files) = std::fs::read_dir(dir) else {
+                warn!(log, "Could not read log directory contents"; "path" => ?dir);
+                break;
+            };
+
+            for file in files {
+                let Ok(dir_entry) = file else {
+                    warn!(log, "Could not read file");
+                    continue;
+                };
+
+                let Ok(file_name) = dir_entry.file_name().into_string() else {
+                    warn!(log, "Could not read file"; "file" => ?dir_entry);
+                    continue;
+                };
+
+                if file_name.starts_with("libp2p.log") | file_name.starts_with("discv5.log") {
+                    let log_file_date = file_name.split('.').collect::<Vec<_>>();
+                    if log_file_date.len() == 3 {
+                        let Ok(log_file_date_type) =
+                            NaiveDate::parse_from_str(log_file_date[2], "%Y-%m-%d")
+                        else {
+                            warn!(log, "Could not parse log file date"; "file" => file_name);
+                            continue;
+                        };
+
+                        if log_file_date_type < yesterday_date.into() {
+                            // Delete the file, its too old
+                            debug!(log, "Removing old log file"; "file" => &file_name);
+                            if let Err(e) = std::fs::remove_file(dir_entry.path()) {
+                                warn!(log, "Failed to remove log file"; "file" => file_name, "error" => %e);
+                            }
+                        }
+                    }
+                }
+            }
+        }
+    }
+}