Skip gc cutoff lsn check at timeline creation if lease exists (#10685)

Right now, branch creation doesn't care if a lsn lease exists or not, it
just fails if the passed lsn is older than either the last or the
planned gc cutoff.

However, if an lsn lease exists for a given lsn, we can actually create
a branch at that point: nothing has been gc'd away.

This prevents race conditions that #10678 still leaves around.

Related: #10639
https://github.com/neondatabase/cloud/issues/23667
This commit is contained in:
Arpad Müller
2025-02-06 11:10:11 +01:00
committed by GitHub
parent b66fbd6176
commit 05326cc247
3 changed files with 22 additions and 15 deletions

View File

@@ -1690,7 +1690,7 @@ impl PageServerHandler {
// to distinguish a misbehaving client (asking for old LSN) from a storage issue (data missing at a legitimate LSN).
if request_lsn < **latest_gc_cutoff_lsn && !timeline.is_gc_blocked_by_lsn_lease_deadline() {
let gc_info = &timeline.gc_info.read().unwrap();
if !gc_info.leases.contains_key(&request_lsn) {
if !gc_info.lsn_covered_by_lease(request_lsn) {
return Err(
PageStreamError::BadRequest(format!(
"tried to request a page version that was garbage collected. requested at {} gc cutoff {}",

View File

@@ -4642,22 +4642,26 @@ impl Tenant {
// check against last actual 'latest_gc_cutoff' first
let latest_gc_cutoff_lsn = src_timeline.get_latest_gc_cutoff_lsn();
src_timeline
.check_lsn_is_in_scope(start_lsn, &latest_gc_cutoff_lsn)
.context(format!(
"invalid branch start lsn: less than latest GC cutoff {}",
*latest_gc_cutoff_lsn,
))
.map_err(CreateTimelineError::AncestorLsn)?;
// and then the planned GC cutoff
{
let gc_info = src_timeline.gc_info.read().unwrap();
let cutoff = gc_info.min_cutoff();
if start_lsn < cutoff {
return Err(CreateTimelineError::AncestorLsn(anyhow::anyhow!(
"invalid branch start lsn: less than planned GC cutoff {cutoff}"
)));
let planned_cutoff = gc_info.min_cutoff();
if gc_info.lsn_covered_by_lease(start_lsn) {
tracing::info!("skipping comparison of {start_lsn} with gc cutoff {} and planned gc cutoff {planned_cutoff} due to lsn lease", *latest_gc_cutoff_lsn);
} else {
src_timeline
.check_lsn_is_in_scope(start_lsn, &latest_gc_cutoff_lsn)
.context(format!(
"invalid branch start lsn: less than latest GC cutoff {}",
*latest_gc_cutoff_lsn,
))
.map_err(CreateTimelineError::AncestorLsn)?;
// and then the planned GC cutoff
if start_lsn < planned_cutoff {
return Err(CreateTimelineError::AncestorLsn(anyhow::anyhow!(
"invalid branch start lsn: less than planned GC cutoff {planned_cutoff}"
)));
}
}
}

View File

@@ -531,6 +531,9 @@ impl GcInfo {
pub(super) fn remove_child_offloaded(&mut self, child_id: TimelineId) -> bool {
self.remove_child_maybe_offloaded(child_id, MaybeOffloaded::Yes)
}
pub(crate) fn lsn_covered_by_lease(&self, lsn: Lsn) -> bool {
self.leases.contains_key(&lsn)
}
}
/// The `GcInfo` component describing which Lsns need to be retained. Functionally, this