Hang diagnostics (#11190)

Release Notes:

- Added diagnostics for main-thread hangs on macOS. These are only
enabled if you've opted into diagnostics.

---------

Co-authored-by: Mikayla <mikayla@zed.dev>
This commit is contained in:
Conrad Irwin 2024-04-29 18:13:28 -06:00 committed by GitHub
parent 04cd8dd0f2
commit 0697b417a0
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
8 changed files with 721 additions and 356 deletions

View file

@ -3,11 +3,10 @@
// Disable command line from opening on release mode
#![cfg_attr(not(debug_assertions), windows_subsystem = "windows")]
mod reliability;
mod zed;
use anyhow::{anyhow, Context as _, Result};
use backtrace::Backtrace;
use chrono::Utc;
use clap::{command, Parser};
use cli::FORCE_CLI_MODE_ENV_VAR_NAME;
use client::{parse_zed_link, telemetry::Telemetry, Client, DevServerToken, UserStore};
@ -19,11 +18,8 @@ use editor::{Editor, EditorMode};
use env_logger::Builder;
use fs::RealFs;
use futures::{future, StreamExt};
use gpui::{
App, AppContext, AsyncAppContext, Context, SemanticVersion, Task, ViewContext, VisualContext,
};
use gpui::{App, AppContext, AsyncAppContext, Context, Task, ViewContext, VisualContext};
use image_viewer;
use isahc::{prelude::Configurable, Request};
use language::LanguageRegistry;
use log::LevelFilter;
@ -31,8 +27,7 @@ use assets::Assets;
use mimalloc::MiMalloc;
use node_runtime::RealNodeRuntime;
use parking_lot::Mutex;
use release_channel::{AppCommitSha, ReleaseChannel, RELEASE_CHANNEL};
use serde::{Deserialize, Serialize};
use release_channel::AppCommitSha;
use settings::{
default_settings, handle_settings_file_changes, watch_config_file, Settings, SettingsStore,
};
@ -40,22 +35,16 @@ use simplelog::ConfigBuilder;
use smol::process::Command;
use std::{
env,
ffi::OsStr,
fs::OpenOptions,
io::{IsTerminal, Write},
panic,
path::Path,
sync::{
atomic::{AtomicU32, Ordering},
Arc,
},
thread,
sync::Arc,
};
use theme::{ActiveTheme, SystemAppearance, ThemeRegistry, ThemeSettings};
use util::{
http::{HttpClient, HttpClientWithUrl},
http::HttpClientWithUrl,
maybe, parse_env_output,
paths::{self, CRASHES_DIR, CRASHES_RETIRED_DIR},
paths::{self},
ResultExt, TryFutureExt,
};
use uuid::Uuid;
@ -93,7 +82,18 @@ fn init_headless(dev_server_token: DevServerToken) {
}
init_logger();
App::new().run(|cx| {
let app = App::new();
let session_id = Uuid::new_v4().to_string();
let (installation_id, _) = app
.background_executor()
.block(installation_id())
.ok()
.unzip();
reliability::init_panic_hook(&app, installation_id.clone(), session_id.clone());
app.run(|cx| {
release_channel::init(env!("CARGO_PKG_VERSION"), cx);
if let Some(build_sha) = option_env!("ZED_COMMIT_SHA") {
AppCommitSha::set_global(AppCommitSha(build_sha.into()), cx);
@ -145,12 +145,7 @@ fn init_headless(dev_server_token: DevServerToken) {
);
handle_settings_file_changes(user_settings_file_rx, cx);
let (installation_id, _) = cx
.background_executor()
.block(installation_id())
.ok()
.unzip();
upload_panics_and_crashes(client.http_client(), installation_id, cx);
reliability::init(client.http_client(), installation_id, cx);
headless::init(
client.clone(),
@ -189,7 +184,7 @@ fn init_ui(args: Args) {
.ok()
.unzip();
let session_id = Uuid::new_v4().to_string();
init_panic_hook(&app, installation_id.clone(), session_id.clone());
reliability::init_panic_hook(&app, installation_id.clone(), session_id.clone());
let git_binary_path = if option_env!("ZED_BUNDLE").as_deref() == Some("true") {
app.path_for_auxiliary_executable("git")
@ -386,7 +381,7 @@ fn init_ui(args: Args) {
cx.set_menus(app_menus());
initialize_workspace(app_state.clone(), cx);
upload_panics_and_crashes(client.http_client(), installation_id, cx);
reliability::init(client.http_client(), installation_id, cx);
cx.activate(true);
@ -688,317 +683,6 @@ fn init_stdout_logger() {
})
.init();
}
#[derive(Serialize, Deserialize)]
struct LocationData {
file: String,
line: u32,
}
#[derive(Serialize, Deserialize)]
struct Panic {
thread: String,
payload: String,
#[serde(skip_serializing_if = "Option::is_none")]
location_data: Option<LocationData>,
backtrace: Vec<String>,
app_version: String,
release_channel: String,
os_name: String,
os_version: Option<String>,
architecture: String,
panicked_on: i64,
#[serde(skip_serializing_if = "Option::is_none")]
installation_id: Option<String>,
session_id: String,
}
#[derive(Serialize)]
struct PanicRequest {
panic: Panic,
}
static PANIC_COUNT: AtomicU32 = AtomicU32::new(0);
fn init_panic_hook(app: &App, installation_id: Option<String>, session_id: String) {
let is_pty = stdout_is_a_pty();
let app_metadata = app.metadata();
panic::set_hook(Box::new(move |info| {
let prior_panic_count = PANIC_COUNT.fetch_add(1, Ordering::SeqCst);
if prior_panic_count > 0 {
// Give the panic-ing thread time to write the panic file
loop {
std::thread::yield_now();
}
}
let thread = thread::current();
let thread_name = thread.name().unwrap_or("<unnamed>");
let payload = info
.payload()
.downcast_ref::<&str>()
.map(|s| s.to_string())
.or_else(|| info.payload().downcast_ref::<String>().map(|s| s.clone()))
.unwrap_or_else(|| "Box<Any>".to_string());
if *release_channel::RELEASE_CHANNEL == ReleaseChannel::Dev {
let location = info.location().unwrap();
let backtrace = Backtrace::new();
eprintln!(
"Thread {:?} panicked with {:?} at {}:{}:{}\n{:?}",
thread_name,
payload,
location.file(),
location.line(),
location.column(),
backtrace,
);
std::process::exit(-1);
}
let app_version = if let Some(version) = app_metadata.app_version {
version.to_string()
} else {
option_env!("CARGO_PKG_VERSION")
.unwrap_or("dev")
.to_string()
};
let backtrace = Backtrace::new();
let mut backtrace = backtrace
.frames()
.iter()
.flat_map(|frame| {
frame
.symbols()
.iter()
.filter_map(|frame| Some(format!("{:#}", frame.name()?)))
})
.collect::<Vec<_>>();
// Strip out leading stack frames for rust panic-handling.
if let Some(ix) = backtrace
.iter()
.position(|name| name == "rust_begin_unwind")
{
backtrace.drain(0..=ix);
}
let panic_data = Panic {
thread: thread_name.into(),
payload,
location_data: info.location().map(|location| LocationData {
file: location.file().into(),
line: location.line(),
}),
app_version: app_version.to_string(),
release_channel: RELEASE_CHANNEL.display_name().into(),
os_name: app_metadata.os_name.into(),
os_version: app_metadata
.os_version
.as_ref()
.map(SemanticVersion::to_string),
architecture: env::consts::ARCH.into(),
panicked_on: Utc::now().timestamp_millis(),
backtrace,
installation_id: installation_id.clone(),
session_id: session_id.clone(),
};
if let Some(panic_data_json) = serde_json::to_string_pretty(&panic_data).log_err() {
log::error!("{}", panic_data_json);
}
if !is_pty {
if let Some(panic_data_json) = serde_json::to_string(&panic_data).log_err() {
let timestamp = chrono::Utc::now().format("%Y_%m_%d %H_%M_%S").to_string();
let panic_file_path = paths::LOGS_DIR.join(format!("zed-{}.panic", timestamp));
let panic_file = std::fs::OpenOptions::new()
.append(true)
.create(true)
.open(&panic_file_path)
.log_err();
if let Some(mut panic_file) = panic_file {
writeln!(&mut panic_file, "{}", panic_data_json).log_err();
panic_file.flush().log_err();
}
}
}
std::process::abort();
}));
}
fn upload_panics_and_crashes(
http: Arc<HttpClientWithUrl>,
installation_id: Option<String>,
cx: &mut AppContext,
) {
let telemetry_settings = *client::TelemetrySettings::get_global(cx);
cx.background_executor()
.spawn(async move {
let most_recent_panic = upload_previous_panics(http.clone(), telemetry_settings)
.await
.log_err()
.flatten();
upload_previous_crashes(http, most_recent_panic, installation_id, telemetry_settings)
.await
.log_err()
})
.detach()
}
/// Uploads panics via `zed.dev`.
async fn upload_previous_panics(
http: Arc<HttpClientWithUrl>,
telemetry_settings: client::TelemetrySettings,
) -> Result<Option<(i64, String)>> {
let panic_report_url = http.build_url("/api/panic");
let mut children = smol::fs::read_dir(&*paths::LOGS_DIR).await?;
let mut most_recent_panic = None;
while let Some(child) = children.next().await {
let child = child?;
let child_path = child.path();
if child_path.extension() != Some(OsStr::new("panic")) {
continue;
}
let filename = if let Some(filename) = child_path.file_name() {
filename.to_string_lossy()
} else {
continue;
};
if !filename.starts_with("zed") {
continue;
}
if telemetry_settings.diagnostics {
let panic_file_content = smol::fs::read_to_string(&child_path)
.await
.context("error reading panic file")?;
let panic: Option<Panic> = serde_json::from_str(&panic_file_content)
.ok()
.or_else(|| {
panic_file_content
.lines()
.next()
.and_then(|line| serde_json::from_str(line).ok())
})
.unwrap_or_else(|| {
log::error!("failed to deserialize panic file {:?}", panic_file_content);
None
});
if let Some(panic) = panic {
most_recent_panic = Some((panic.panicked_on, panic.payload.clone()));
let body = serde_json::to_string(&PanicRequest { panic }).unwrap();
let request = Request::post(&panic_report_url)
.redirect_policy(isahc::config::RedirectPolicy::Follow)
.header("Content-Type", "application/json")
.body(body.into())?;
let response = http.send(request).await.context("error sending panic")?;
if !response.status().is_success() {
log::error!("Error uploading panic to server: {}", response.status());
}
}
}
// We've done what we can, delete the file
std::fs::remove_file(child_path)
.context("error removing panic")
.log_err();
}
Ok::<_, anyhow::Error>(most_recent_panic)
}
static LAST_CRASH_UPLOADED: &'static str = "LAST_CRASH_UPLOADED";
/// upload crashes from apple's diagnostic reports to our server.
/// (only if telemetry is enabled)
async fn upload_previous_crashes(
http: Arc<HttpClientWithUrl>,
most_recent_panic: Option<(i64, String)>,
installation_id: Option<String>,
telemetry_settings: client::TelemetrySettings,
) -> Result<()> {
if !telemetry_settings.diagnostics {
return Ok(());
}
let last_uploaded = KEY_VALUE_STORE
.read_kvp(LAST_CRASH_UPLOADED)?
.unwrap_or("zed-2024-01-17-221900.ips".to_string()); // don't upload old crash reports from before we had this.
let mut uploaded = last_uploaded.clone();
let crash_report_url = http.build_zed_api_url("/telemetry/crashes", &[])?;
// crash directories are only set on MacOS
for dir in [&*CRASHES_DIR, &*CRASHES_RETIRED_DIR]
.iter()
.filter_map(|d| d.as_deref())
{
let mut children = smol::fs::read_dir(&dir).await?;
while let Some(child) = children.next().await {
let child = child?;
let Some(filename) = child
.path()
.file_name()
.map(|f| f.to_string_lossy().to_lowercase())
else {
continue;
};
if !filename.starts_with("zed-") || !filename.ends_with(".ips") {
continue;
}
if filename <= last_uploaded {
continue;
}
let body = smol::fs::read_to_string(&child.path())
.await
.context("error reading crash file")?;
let mut request = Request::post(&crash_report_url.to_string())
.redirect_policy(isahc::config::RedirectPolicy::Follow)
.header("Content-Type", "text/plain");
if let Some((panicked_on, payload)) = most_recent_panic.as_ref() {
request = request
.header("x-zed-panicked-on", format!("{}", panicked_on))
.header("x-zed-panic", payload)
}
if let Some(installation_id) = installation_id.as_ref() {
request = request.header("x-zed-installation-id", installation_id);
}
let request = request.body(body.into())?;
let response = http.send(request).await.context("error sending crash")?;
if !response.status().is_success() {
log::error!("Error uploading crash to server: {}", response.status());
}
if uploaded < filename {
uploaded = filename.clone();
KEY_VALUE_STORE
.write_kvp(LAST_CRASH_UPLOADED.to_string(), filename)
.await?;
}
}
}
Ok(())
}
async fn load_login_shell_environment() -> Result<()> {
let marker = "ZED_LOGIN_SHELL_START";
let shell = env::var("SHELL").context(