compute: add non-skippable apply config operations.

apply_config() step of compute start is controlled by skip_pg_catalog_updates flag,
this is a performance optimization to decrease compute startup time, but it introduces extra dependency on cplane.
Introduce small subset of operations that we run always, independent from this flag.
This commit is contained in:
Anastasia Lubennikova
2025-03-21 15:13:11 +00:00
parent ff87b51300
commit 15b06a6a1e
2 changed files with 106 additions and 0 deletions

View File

@@ -1497,6 +1497,27 @@ impl ComputeNode {
Ok::<(), anyhow::Error>(())
}
/// Apply config operations that are not covered by `skip_pg_catalog_updates`
#[instrument(skip_all)]
pub fn apply_config_non_skippable(&self, compute_state: &ComputeState) -> Result<()> {
let conf = self.get_tokio_conn_conf(Some("compute_ctl:apply_config"));
let conf = Arc::new(conf);
let spec = Arc::new(
compute_state
.pspec
.as_ref()
.expect("spec must be set")
.spec
.clone(),
);
// Merge-apply spec & changes to PostgreSQL state.
self.apply_spec_sql_non_skippable(spec.clone(), conf.clone())?;
Ok::<(), anyhow::Error>(())
}
// Wrapped this around `pg_ctl reload`, but right now we don't use
// `pg_ctl` for start / stop.
#[instrument(skip_all)]
@@ -1619,8 +1640,24 @@ impl ComputeNode {
"updated postgresql.conf to set neon.disable_logical_replication_subscribers=false"
);
}
self.pg_reload_conf()?;
} else {
// We need to run some operations even if skip_pg_catalog_updates is set
let pgdata_path = Path::new(&self.params.pgdata);
// temporarily reset max_cluster_size in config
// to avoid the possibility of hitting the limit, while we are applying config:
// creating new extensions, roles, etc...
config::with_compute_ctl_tmp_override(pgdata_path, "neon.max_cluster_size=-1", || {
self.pg_reload_conf()?;
self.apply_config_non_skippable(compute_state)?;
Ok(())
})?;
self.pg_reload_conf()?;
}
self.post_apply_config()?;
Ok(())

View File

@@ -308,6 +308,75 @@ impl ComputeNode {
Ok(())
}
// Similar to apply_spec_sql, but for the simplified set of operations
// that we perform even when `pg_skip_catalog_updates` is set.
//
// Keep the list of operations as small as possible,
// as it will be run on every spec change and affect compute start time.
pub fn apply_spec_sql_non_skippable(
&self,
spec: Arc<ComputeSpec>,
conf: Arc<tokio_postgres::Config>,
) -> Result<()> {
info!("Applying non_skippable config",);
debug!("Config: {:?}", spec);
let rt = tokio::runtime::Handle::current();
rt.block_on(async {
let client = Self::get_maintenance_client(&conf).await?;
let spec = spec.clone();
let jwks_roles = Arc::new(
spec.as_ref()
.local_proxy_config
.iter()
.flat_map(|it| &it.jwks)
.flatten()
.flat_map(|setting| &setting.role_names)
.cloned()
.collect::<HashSet<_>>(),
);
// NOTE: Here we assume that operations below don't use ctx
// TODO: refactor apply_operations() to accept ctx as option.
let ctx = Arc::new(tokio::sync::RwLock::new(MutableApplyContext {
roles: HashMap::new(),
dbs: HashMap::new(),
}));
let mut phases = vec![];
match spec.audit_log_level {
ComputeAudit::Hipaa => {
phases.push(CreatePgauditExtension);
phases.push(CreatePgauditlogtofileExtension);
phases.push(DisablePostgresDBPgAudit);
}
ComputeAudit::Log => {
phases.push(CreatePgauditExtension);
phases.push(DisablePostgresDBPgAudit);
}
ComputeAudit::Disabled => {}
}
for phase in phases {
debug!("Applying phase {:?}", &phase);
apply_operations(
spec.clone(),
ctx.clone(),
jwks_roles.clone(),
phase,
|| async { Ok(&client) },
)
.await?;
}
Ok::<(), anyhow::Error>(())
})?;
Ok(())
}
/// Apply SQL migrations of the RunInEachDatabase phase.
///
/// May opt to not connect to databases that don't have any scheduled