From e288bfe8f036d995d74367acef4b2fa0f04ecf26 Mon Sep 17 00:00:00 2001 From: Matthew Gumport Date: Wed, 4 Sep 2024 16:02:29 -0700 Subject: [PATCH] change lock status memory ordering, tag spans This changes the memory ordering for the lock status load to `SeqCst` from `Relaxed` to eliminate a potential source of panics. Panics had the frames: ``` pingora_proxy::proxy_cache::::handle_lock_status (proxy_cache.rs:748) pingora_proxy::proxy_cache::::proxy_cache::{{closure}} (proxy_cache.rs:211) pingora_proxy::HttpProxy::process_request::{{closure}} (lib.rs:509) pingora_proxy::HttpProxy::process_new_http::{{closure}} (lib.rs:727) ``` which showed we were checking on the status of the lock, after waiting on it, and still seeing its status as waiting. The status is returned by value, so this is not a time-of-check to time-of-use problem, this is an inconsistency in how the lock status is managed. The change in memory order is mostly for the sake of this programmer's attempts to understand what is happening. This also completes a couple of TODOs to limit the wait period as well as tag the span with the lock status. --- .bleep | 2 +- pingora-cache/Cargo.toml | 1 + pingora-cache/src/lib.rs | 8 ++++++-- pingora-cache/src/lock.rs | 27 ++++++++++++++++++++------- 4 files changed, 28 insertions(+), 10 deletions(-) diff --git a/.bleep b/.bleep index 0f44214..5081a5b 100644 --- a/.bleep +++ b/.bleep @@ -1 +1 @@ -30c624970d46b07efd110fcfb8dd0b6f9a099e2b \ No newline at end of file +2351cdf592f9986201d754e6ee1f37f493f69abb \ No newline at end of file diff --git a/pingora-cache/Cargo.toml b/pingora-cache/Cargo.toml index 2dc3f2f..6ae6c12 100644 --- a/pingora-cache/Cargo.toml +++ b/pingora-cache/Cargo.toml @@ -43,6 +43,7 @@ lru = { workspace = true } ahash = { workspace = true } hex = "0.4" httparse = { workspace = true } +strum = { version = "0.26", features = ["derive"] } [dev-dependencies] tokio-test = "0.4" diff --git a/pingora-cache/src/lib.rs b/pingora-cache/src/lib.rs index cdcfcc9..f1f8bd2 100644 --- a/pingora-cache/src/lib.rs +++ b/pingora-cache/src/lib.rs @@ -21,6 +21,7 @@ use key::{CacheHashKey, HashBinary}; use lock::WritePermit; use pingora_error::Result; use pingora_http::ResponseHeader; +use rustracing::tag::Tag; use std::time::{Duration, Instant, SystemTime}; use trace::CacheTraceCTX; @@ -1047,7 +1048,7 @@ impl HttpCache { /// Check [Self::is_cache_locked()], panic if this request doesn't have a read lock. pub async fn cache_lock_wait(&mut self) -> LockStatus { let inner = self.inner_mut(); - let _span = inner.traces.child("cache_lock"); + let mut span = inner.traces.child("cache_lock"); let lock = inner.lock.take(); // remove the lock from self if let Some(Locked::Read(r)) = lock { let now = Instant::now(); @@ -1059,7 +1060,10 @@ impl HttpCache { .lock_duration .map_or(lock_duration, |d| d + lock_duration), ); - r.lock_status() // TODO: tag the span with lock status + let status = r.lock_status(); + let tag_value: &'static str = status.into(); + span.set_tag(|| Tag::new("status", tag_value)); + status } else { // should always call is_cache_locked() before this function panic!("cache_lock_wait on wrong type of lock") diff --git a/pingora-cache/src/lock.rs b/pingora-cache/src/lock.rs index 7f50691..8853cb1 100644 --- a/pingora-cache/src/lock.rs +++ b/pingora-cache/src/lock.rs @@ -100,12 +100,14 @@ impl CacheLock { } } +use log::warn; use std::sync::atomic::{AtomicU8, Ordering}; use std::time::{Duration, Instant}; +use strum::IntoStaticStr; use tokio::sync::Semaphore; /// Status which the read locks could possibly see. -#[derive(Debug, Copy, Clone, PartialEq, Eq)] +#[derive(Debug, Copy, Clone, PartialEq, Eq, IntoStaticStr)] pub enum LockStatus { /// Waiting for the writer to populate the asset Waiting, @@ -180,7 +182,7 @@ impl LockCore { } fn lock_status(&self) -> LockStatus { - self.lock_status.load(Ordering::Relaxed).into() + self.lock_status.load(Ordering::SeqCst).into() } } @@ -197,11 +199,22 @@ impl ReadLock { return; } - // TODO: should subtract now - start so that the lock don't wait beyond start + timeout - // Also need to be careful not to wake everyone up at the same time + // TODO: need to be careful not to wake everyone up at the same time // (maybe not an issue because regular cache lock release behaves that way) - let _ = timeout(self.0.timeout, self.0.lock.acquire()).await; - // permit is returned to Semaphore right away + if let Some(duration) = self.0.timeout.checked_sub(self.0.lock_start.elapsed()) { + match timeout(duration, self.0.lock.acquire()).await { + Ok(Ok(_)) => { // permit is returned to Semaphore right away + } + Ok(Err(e)) => { + warn!("error acquiring semaphore {e:?}") + } + Err(_) => { + self.0 + .lock_status + .store(LockStatus::Timeout.into(), Ordering::SeqCst); + } + } + } } /// Test if it is still locked @@ -211,7 +224,7 @@ impl ReadLock { /// Whether the lock is expired, e.g., the writer has been holding the lock for too long pub fn expired(&self) -> bool { - // NOTE: this whether the lock is currently expired + // NOTE: this is whether the lock is currently expired // not whether it was timed out during wait() self.0.lock_start.elapsed() >= self.0.timeout }