Event Search, Postgres Queue & Teams Alerts

Here's what shipped in 26.7.3.

Find an event without scrolling

The Events log has a search box now. Type part of an event type, a source name, an idempotency key, or the start of an event ID, and you get the matching events from the date range you already have open.

Paste JSON instead and Convoy searches the payload itself, so you can pull up every event that carried a particular customer, invoice, or amount:

{"event": "invoice.paid"}

Everything you paste has to match in the same place in the body, so a top-level field is not the same as one nested under data. Quotes are optional, and you can mix text with JSON when you want both:

invoice.paid {amount: 1200}

Open a result and Convoy highlights what you searched for in the event body, so you can see why it matched.

Event search is a Business plan feature. Without it the box stays where it is, greyed out and tagged Business, and the API answers 403 if you pass query or body.

Run the queue on Postgres, without Redis (experimental)

If you would rather not operate Redis, the delivery queue can now live in Postgres. Set queue_provider to postgres, or CONVOY_QUEUE_PROVIDER if you configure by environment, and caching, rate limits, and circuit breaker state move across with it. Postgres becomes the only thing you have to run.

This one is experimental and Redis is still the default. You need a license that includes it and the postgres-queue feature flag, and Convoy refuses to boot without both rather than quietly falling back to Redis. Point every server and agent at the same provider, or your writers and your workers end up on different queues.

Failure alerts in Microsoft Teams

Endpoint failure notifications can go to a Teams channel, alongside Slack and email. Paste the Teams webhook URL on the endpoint, and make it a Workflows (Power Automate) URL: the retired Office 365 connector URLs no longer deliver. Convoy only alerts when an endpoint actually changes state, so an endpoint that is already down does not keep buzzing the channel. Same Advanced Endpoint Management license as the other channels, and without that license the field is dropped rather than saved.

The dashboard stops under-reporting your numbers

The summary numbers on the project overview used to be counted from scratch on every load. Big projects outran the clock, and a count that gave up showed as 0. You could be looking at zero successful and zero failed next to hundreds of thousands of events sent.

Convoy now keeps a daily running total in the background and reads that instead. The page costs the same no matter how much history you keep, and when something does go wrong you get a dash rather than a wrong zero. The Prometheus gauges you already scrape read those totals too, under the same names.

Two things to expect. Today's numbers can sit up to a minute behind, and just after you upgrade the older days fill in as Convoy works back through your history.

A query that is too wide says so

Ask Event Deliveries for too much at once and you get a 504 with Event deliveries took too long. Narrow the date range. instead of a request that hangs and then fails without telling you why. Narrow the window, or filter by endpoint.

Broken indexes fix themselves

Postgres can leave an index half-built after a failed upgrade or an interrupted migration. Until now it sat there, invalid and ignored, quietly making your queries slow until someone noticed. Convoy now picks those up at startup and rebuilds them in the background, unique indexes first. Admin → Table indexes shows what is left, and Rebuild retries anything that failed.

Breaking: rebuilding an index means dropping it first. If the broken one was a unique index, it stops enforcing uniqueness until the rebuild finishes. Upgrade in a window where that is acceptable, and watch that page until everything is valid again. If a rebuild runs into rows that are already duplicated, Convoy stops and shows you the key involved rather than retrying forever, and those duplicates need clearing by hand.

Smaller fixes

• Event Deliveries loads much faster on projects using retention partitions
• Delivery details show the URL that attempt actually called
• The Events log spinner waits for your search to finish, and the deliveries panel fills in for the event you picked instead of waiting for a click
• Admin → Table partitions offers the operation that fits the table you just selected
• Prometheus no longer counts deliveries while a process is still starting up
• Verify-email matches the sign-in page, and refreshing after a successful verify no longer shows an error
• The local install script pulls the published Convoy image instead of building one

Open the Events log and try a JSON filter.