Upgrade to v1.8

This guide covers what changes when upgrading from v1.7 to v1.8. There is one breaking change, one ClickHouse migration and four behavior changes to review before upgrading.

With the ClickHouse log store, run outpost migrate apply --yes before starting v1.8: the server refuses to start while a migration is pending. There is no new PostgreSQL migration. Retries scheduled by v1.7 run normally on v1.8, and you can roll back to v1.7.0 without reverting the migration.

ChangeImpactAction Required
Webhook deliveries no longer follow redirectsWebhook destinations whose URL answers with a redirectUpdate the destination URL to the final address

For new features and fixes, see the v1.8.0 release notes.

Breaking Changes

Webhook deliveries no longer follow redirects

A 3xx response is now a failed attempt with its status code, retried like any other failure and, when enabled, counted toward auto-disable. v1.7 followed up to 10 redirects. On a 301, 302 or 303 it re-sent the request as a GET without the body, so the new URL never received the event, yet the attempt was recorded as a success.

Common causes are an http:// URL redirected to https:// and a missing or extra trailing slash. The failed attempt's response data includes the redirect target in location, so you can see the URL to switch to. The redirect_error error code is removed, as it can no longer occur. See Webhook.

Behavior Changes

  • With CLICKHOUSE_LOG_RETENTION_TTL_DAYS set, ClickHouse removes expired logs a whole part at a time instead of rewriting parts to delete expired rows. Tables are partitioned by month, so logs can stay queryable and on disk for up to about a month after their retention ends. With a short retention, plan for the extra disk.
  • Hookdeck destinations retry a failed delivery. v1.7 recorded a non-2xx response from Hookdeck as failed but never retried it, and treated a 3xx as a success. Each retry is a new failed attempt, so a failing Hookdeck destination reaches consecutive failure alerts and auto-disable sooner, and can now send alert.attempt.exhausted_retries.
  • A receiver that closes the connection before responding is recorded as connection_reset. v1.7 recorded it as network_error, which happened most often through a forward proxy (DESTINATIONS_PROXY_URL).
  • With auto-disable off, the 100% alert.destination.consecutive_failure alert is sent once per failure streak. v1.7 sent it again on every failure past the threshold.

New Configuration

See the configuration reference for defaults and details.