Skip to content

v1.8.0

Choose a tag to compare

@alexluong alexluong released this 10 Nov 10:14
· 37 commits to main since this release
c0a8457

⚠️ Breaking Changes

See the v1.8 Upgrade Guide for details.

  • Webhook deliveries no longer follow redirects. A 3xx response is a failed attempt, retried like any other failure, and the redirect target is recorded in the attempt's response data as location. Before, a 301, 302 or 303 was followed as a GET without the body and recorded as a success, though the event never arrived. Update destinations that answer with a redirect, such as http:// to https:// or a missing trailing slash, to their final URL. by @alexluong in #1175

Features

  • DESTINATIONS_BLOCK_PRIVATE_ADDRESSES keeps webhook destinations on public addresses: a private URL gets 422 on create and update, and a delivery that resolves to a private address fails with address_not_allowed, including after a DNS change. DESTINATIONS_PRIVATE_ADDRESS_ALLOWLIST lists IPs and ranges that stay reachable. Off by default. by @alexluong in #1176
  • Lighter ClickHouse log expiry: expired logs are dropped in whole parts instead of rewritten, which cuts CPU spikes on large tables. Logs can stay up to about a month past their retention. Requires a ClickHouse migration (outpost migrate apply). by @alexluong in #1178
  • Attempt lists read event data, event metadata and response data from the log store only when include asks for them. Responses are unchanged. by @alexluong in #1187

Refactors

  • Destination types can mark a failure as not retryable or leave it out of the consecutive-failure count, and bring their own retry schedule. No existing destination type uses it, so delivery behavior is unchanged. by @alexluong in #1174

Fixes

  • Hookdeck destinations retry failed deliveries. Before, a non-2xx response from Hookdeck was recorded as failed but never retried, and a 3xx was recorded as a success. by @Arunendra21 in #1202
  • Concurrent writes to the same tenant or destination no longer corrupt each other: a create can't overwrite an existing destination or exceed the tenant's limit, an update can't bring back a deleted destination, and concurrent updates no longer overwrite each other. After 3 conflicting updates in a row the API returns 409. by @alexluong in #1172
  • Stale ClickHouse connections after a ClickHouse restart behind a proxy or load balancer are replaced instead of failing calls one at a time, and reads retry once on a fresh connection. Idle connection pools close after CLICKHOUSE_IDLE_CLOSE_SECONDS (default 300). by @alexluong in #1186
  • The 100% alert.destination.consecutive_failure alert is sent once per failure streak. With auto-disable off, it used to repeat on every failure past the threshold. by @alexluong in #1197
  • Operator events sent to AWS SQS or GCP Pub/Sub now go out. Those sinks were never initialized, so Outpost crashed when it sent its first operator event. All queue sinks now connect at startup, so a misconfigured sink stops Outpost from starting instead of failing later. by @hogklint in #1196
  • Retries running when Outpost shuts down finish before it stops, instead of being delayed or delivered twice after the restart. by @alexluong in #1185
  • A receiver that closes the connection before responding is recorded as connection_reset instead of network_error, which was most common through a forward proxy. by @alexluong in #1191
  • Shutdown and concurrency fixes: the log consumer's shutdown could hang, a consumer could hold a message it couldn't run yet and leave it unacked on shutdown, and a publish could start on a closed publisher. by @alexluong in #1160, #1166, #1169
  • Webhook response bodies are closed after each delivery, so a response over the size limit no longer holds its connection open. by @alexluong in #1206

Updates

New Contributors

Full Changelog: v1.7.0...v1.8.0