Compare commits

..

6 Commits

Author SHA1 Message Date
John Spray
550eadc1d5 pageserver: cargo fmt
This is in a separate commit to make the previous ones more readable.
2023-08-03 14:12:46 +01:00
John Spray
db0deb8457 pageserver: respect task_mgr cancellation in metrics task
This previously relied on seeing a channel close, when
Tenant is destroyed: this task ran beyond Tenant::shutdown,
whereas the idea of that shutdown function is that all the
per-tenant background tasks are joined when it completes.

Instead, stop this task as soon as background tasks for the
Tenant are cancelled, making the behavior of shutdown() much
more obvious and bringing the task into line with how we
do shutdown in other background tasks.
2023-08-03 10:01:12 +01:00
John Spray
a7ad080961 pageserver: remove bare tokio::spawn
Commit ddb9c2fe9 added this task, which is launched
with tokio::spawn rather than task_mgr::spawn, and is
not wrapped in a tracing Instrumented<>.

While the task doesn't overtly do any logging, for
consistency we should spawn all our tasks via the wrapper: in
future this task might be extended to e.g. emit log lines that
we would expect to have the proper tracing spans.
2023-08-03 09:53:38 +01:00
John Spray
df49a9b7aa pagekeeper: suppress error logs in shutdown/detach (#4876)
## Problem

Error messages like this coming up during normal operations:
```
        Compaction failed, retrying in 2s: timeline is Stopping

       Compaction failed, retrying in 2s: Cannot run compaction iteration on inactive tenant
```

## Summary of changes

Add explicit handling for the shutdown case in these locations, to
suppress error logs.
2023-08-02 19:31:09 +01:00
bojanserafimov
4ad0c8f960 compute_ctl: Prewarm before starting http server (#4867) 2023-08-02 14:19:06 -04:00
Joonas Koivunen
e0b05ecafb build: ca-certificates need to be present (#4880)
as needed since #4715 or this will happen:

```
ERROR panic{thread=main location=.../hyper-rustls-0.23.2/src/config.rs:48:9}: no CA certificates found
```
2023-08-02 20:34:21 +03:00
6 changed files with 82 additions and 50 deletions

View File

@@ -816,6 +816,7 @@ COPY --from=compute-tools --chown=postgres /home/nonroot/target/release-line-deb
# libxml2, libxslt1.1 for xml2
# libzstd1 for zstd
# libboost*, libfreetype6, and zlib1g for rdkit
# ca-certificates for communicating with s3 by compute_ctl
RUN apt update && \
apt install --no-install-recommends -y \
gdb \
@@ -839,7 +840,8 @@ RUN apt update && \
libcurl4-openssl-dev \
locales \
procps \
zlib1g && \
zlib1g \
ca-certificates && \
rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* && \
localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8

View File

@@ -203,6 +203,14 @@ fn main() -> Result<()> {
};
let compute = Arc::new(compute_node);
// If this is a pooled VM, prewarm before starting HTTP server and becoming
// available for binding. Prewarming helps postgres start quicker later,
// because QEMU will already have it's memory allocated from the host, and
// the necessary binaries will alreaady be cached.
if !spec_set {
compute.prewarm_postgres()?;
}
// Launch http service first, so we were able to serve control-plane
// requests, while configuration is still in progress.
let _http_handle =
@@ -214,12 +222,6 @@ fn main() -> Result<()> {
// No spec provided, hang waiting for it.
info!("no compute spec provided, waiting");
// TODO this can stall startups in the unlikely event that we bind
// this compute node while it's busy prewarming. It's not too
// bad because it's just 100ms and unlikely, but it's an
// avoidable problem.
compute.prewarm_postgres()?;
let mut state = compute.state.lock().unwrap();
while state.status != ComputeStatus::ConfigurationPending {
state = compute.state_changed.wait(state).unwrap();

View File

@@ -2076,57 +2076,74 @@ impl Tenant {
) -> Tenant {
let (state, mut rx) = watch::channel(state);
tokio::spawn(async move {
let tid = tenant_id.to_string();
task_mgr::spawn(
task_mgr::BACKGROUND_RUNTIME.handle(),
TaskKind::MetricsCollection,
Some(tenant_id),
None,
&format!("state metrics collector for tenant {tenant_id}"),
false,
async move {
let cancel = task_mgr::shutdown_token();
fn inspect_state(state: &TenantState) -> ([&'static str; 1], bool) {
([state.into()], matches!(state, TenantState::Broken { .. }))
}
let tid = tenant_id.to_string();
let mut tuple = inspect_state(&rx.borrow_and_update());
let is_broken = tuple.1;
let mut counted_broken = if !is_broken {
// the tenant might be ignored and reloaded, so first remove any previous set
// element. it most likely has already been scraped, as these are manual operations
// right now. most likely we will add it back very soon.
drop(crate::metrics::BROKEN_TENANTS_SET.remove_label_values(&[&tid]));
false
} else {
// add the id to the set right away, there should not be any updates on the channel
// after
crate::metrics::BROKEN_TENANTS_SET
.with_label_values(&[&tid])
.set(1);
true
};
loop {
let labels = &tuple.0;
let current = TENANT_STATE_METRIC.with_label_values(labels);
current.inc();
if rx.changed().await.is_err() {
// tenant has been dropped; decrement the counter because a tenant with that
// state is no longer in tenant map, but allow any broken set item to exist
// still.
current.dec();
break;
fn inspect_state(state: &TenantState) -> ([&'static str; 1], bool) {
([state.into()], matches!(state, TenantState::Broken { .. }))
}
current.dec();
tuple = inspect_state(&rx.borrow_and_update());
let mut tuple = inspect_state(&rx.borrow_and_update());
let is_broken = tuple.1;
if is_broken && !counted_broken {
counted_broken = true;
// insert the tenant_id (back) into the set
let mut counted_broken = if !is_broken {
// the tenant might be ignored and reloaded, so first remove any previous set
// element. it most likely has already been scraped, as these are manual operations
// right now. most likely we will add it back very soon.
drop(crate::metrics::BROKEN_TENANTS_SET.remove_label_values(&[&tid]));
false
} else {
// add the id to the set right away, there should not be any updates on the channel
// after
crate::metrics::BROKEN_TENANTS_SET
.with_label_values(&[&tid])
.inc();
.set(1);
true
};
loop {
let labels = &tuple.0;
let current = TENANT_STATE_METRIC.with_label_values(labels);
current.inc();
let changed = tokio::select! {
changed = rx.changed() => {changed},
_ = cancel.cancelled() => {return Ok(())}
};
if changed.is_err() {
// tenant has been dropped; decrement the counter because a tenant with that
// state is no longer in tenant map, but allow any broken set item to exist
// still.
current.dec();
break;
}
current.dec();
tuple = inspect_state(&rx.borrow_and_update());
let is_broken = tuple.1;
if is_broken && !counted_broken {
counted_broken = true;
// insert the tenant_id (back) into the set
crate::metrics::BROKEN_TENANTS_SET
.with_label_values(&[&tid])
.inc();
}
}
Ok(())
}
});
.instrument(info_span!("state_metrics", tenant_id = %tenant_id)),
);
Tenant {
tenant_id,

View File

@@ -103,6 +103,11 @@ async fn compaction_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
}
}
if cancel.is_cancelled() {
info!("received cancellation request");
break;
}
let started_at = Instant::now();
let sleep_duration = if period == Duration::ZERO {

View File

@@ -697,6 +697,9 @@ impl Timeline {
Err(CompactionError::DownloadRequired(rls)) => {
anyhow::bail!("Compaction requires downloading multiple times (last was {} layers), possibly battling against eviction", rls.len())
}
Err(CompactionError::ShuttingDown) => {
return Ok(());
}
Err(CompactionError::Other(e)) => {
return Err(e);
}
@@ -778,7 +781,8 @@ impl Timeline {
let layer_removal_cs = Arc::new(self.layer_removal_cs.clone().lock_owned().await);
// Is the timeline being deleted?
if self.is_stopping() {
return Err(anyhow::anyhow!("timeline is Stopping").into());
trace!("Dropping out of compaction on timeline shutdown");
return Err(CompactionError::ShuttingDown);
}
let target_file_size = self.get_checkpoint_distance();
@@ -3235,6 +3239,8 @@ enum CompactionError {
/// This should not happen repeatedly, but will be retried once by top-level
/// `Timeline::compact`.
DownloadRequired(Vec<Arc<RemoteLayer>>),
/// The timeline or pageserver is shutting down
ShuttingDown,
/// Compaction cannot be done right now; page reconstruction and so on.
Other(anyhow::Error),
}

View File

@@ -13,7 +13,7 @@ pub struct PasswordHackPayload {
impl PasswordHackPayload {
pub fn parse(bytes: &[u8]) -> Option<Self> {
// The format is `project=<utf-8>;<password-bytes>`.
let mut iter = bytes.splitn_str(2, &[';', '$']);
let mut iter = bytes.splitn_str(2, ";");
let endpoint = iter.next()?.to_str().ok()?;
let endpoint = parse_endpoint_param(endpoint)?.to_owned();
let password = iter.next()?.to_owned();