storcon: signal LSN wait to pageserver during live migration (#10452)

## Problem

We've seen the ingest connection manager get stuck shortly after a
migration.

## Summary of changes

A speculative mitigation is to use the same mechanism as get page
requests for kicking LSN ingest. The connection manager monitors
LSN waits and queries the broker if no updates are received for the
timeline.

Closes https://github.com/neondatabase/neon/issues/10351
This commit is contained in:
Vlad Lazar
2025-01-28 17:33:07 +00:00
committed by GitHub
parent 1010b8add4
commit c54cd9e76a
10 changed files with 193 additions and 12 deletions
+64
View File
@@ -10,6 +10,7 @@ use std::time::Duration;
use anyhow::{anyhow, Context, Result};
use enumset::EnumSet;
use futures::future::join_all;
use futures::StreamExt;
use futures::TryFutureExt;
use humantime::format_rfc3339;
@@ -40,6 +41,7 @@ use pageserver_api::models::TenantShardSplitRequest;
use pageserver_api::models::TenantShardSplitResponse;
use pageserver_api::models::TenantSorting;
use pageserver_api::models::TenantState;
use pageserver_api::models::TenantWaitLsnRequest;
use pageserver_api::models::TimelineArchivalConfigRequest;
use pageserver_api::models::TimelineCreateRequestMode;
use pageserver_api::models::TimelineCreateRequestModeImportPgdata;
@@ -95,6 +97,8 @@ use crate::tenant::timeline::CompactOptions;
use crate::tenant::timeline::CompactRequest;
use crate::tenant::timeline::CompactionError;
use crate::tenant::timeline::Timeline;
use crate::tenant::timeline::WaitLsnTimeout;
use crate::tenant::timeline::WaitLsnWaiter;
use crate::tenant::GetTimelineError;
use crate::tenant::OffloadedTimeline;
use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
@@ -2790,6 +2794,63 @@ async fn secondary_download_handler(
json_response(status, progress)
}
async fn wait_lsn_handler(
mut request: Request<Body>,
cancel: CancellationToken,
) -> Result<Response<Body>, ApiError> {
let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
let wait_lsn_request: TenantWaitLsnRequest = json_request(&mut request).await?;
let state = get_state(&request);
let tenant = state
.tenant_manager
.get_attached_tenant_shard(tenant_shard_id)?;
let mut wait_futures = Vec::default();
for timeline in tenant.list_timelines() {
let Some(lsn) = wait_lsn_request.timelines.get(&timeline.timeline_id) else {
continue;
};
let fut = {
let timeline = timeline.clone();
let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
async move {
timeline
.wait_lsn(
*lsn,
WaitLsnWaiter::HttpEndpoint,
WaitLsnTimeout::Custom(wait_lsn_request.timeout),
&ctx,
)
.await
}
};
wait_futures.push(fut);
}
if wait_futures.is_empty() {
return json_response(StatusCode::NOT_FOUND, ());
}
let all_done = tokio::select! {
results = join_all(wait_futures) => {
results.iter().all(|res| res.is_ok())
},
_ = cancel.cancelled() => {
return Err(ApiError::Cancelled);
}
};
let status = if all_done {
StatusCode::OK
} else {
StatusCode::ACCEPTED
};
json_response(status, ())
}
async fn secondary_status_handler(
request: Request<Body>,
_cancel: CancellationToken,
@@ -3577,6 +3638,9 @@ pub fn make_router(
.post("/v1/tenant/:tenant_shard_id/secondary/download", |r| {
api_handler(r, secondary_download_handler)
})
.post("/v1/tenant/:tenant_shard_id/wait_lsn", |r| {
api_handler(r, wait_lsn_handler)
})
.put("/v1/tenant/:tenant_shard_id/break", |r| {
testing_api_handler("set tenant state to broken", r, handle_tenant_break)
})