Skip to content
HookDeploy
← All posts

Reliability4 min read

The Outage Ends. The Cleanup Doesn't.

Why HookDeploy exists: a firsthand account of cleaning up after Azure and AWS outages, hunting down which webhooks made it and which silently didn't, and why capturing and replaying webhooks beats manual reconciliation every time.

Michael Bernstein

Founder, HookDeploy

Before HookDeploy, I spent years leading business operations across a handful of industries, and somewhere along the way I became the default IT person, the networking person, the VOIP person, the integration specialist, the guy who builds a custom solution when no single platform actually solves the problem. Sometimes that meant standing up my own hardware. Sometimes it meant stitching things together through Zapier or Make. But almost every one of those problems had the same shape underneath: something happens in System A, and System B needs to know about it, right now, reliably. That’s a webhook. It’s always been a webhook.

Nobody treats that trigger layer like it matters. Until it fails, and then it’s the only thing that matters.

What cleanup actually looks like

I’ve now sat through three separate major cloud outages and done the cleanup afterward: Azure’s outage in July 2024, Azure Front Door going down again in October 2025, and AWS US-East-1’s DNS failure that same October. Three different providers, three different root causes, all within about fifteen months of each other.

The outage itself is never the hard part. Providers post a status update, things come back, everyone moves on. The hard part starts after. You’re not looking at a clean “everything failed” or “everything worked.” You’re looking at a partial, messy, unknowable window where some webhooks fired and made it through, some fired and got dropped, and some never fired at all because the upstream service itself was struggling. Nobody hands you a list. You have to reconstruct it.

That reconstruction is what actually eats the days. Cross-referencing order records against payment records against fulfillment records, trying to figure out which side has the truth. Manually re-entering data for events that never arrived. Trying to simulate a webhook firing again after the fact, hoping you’re recreating the payload accurately enough that downstream systems don’t choke on it or double-process something that actually did make it through the first time. It’s slow, it’s error-prone, and it’s the kind of work that doesn’t show up on a roadmap because it’s not something you planned for. It’s something that happened to you.

The pattern nobody wants to admit

Three major outages in fifteen months, across two different providers, is not a fluke. It’s the actual reliability profile of depending on any single cloud provider for the thing your business runs on. AWS US-East-1 alone logged ten separate outages in 2025 with almost 34 hours of cumulative downtime. That’s not a rare event you insure against just in case. That’s a recurring cost you’re already paying, whether or not you’ve built anything to handle it.

Most businesses don’t have a webhook problem until the day they very suddenly do. And on that day, the actual damage isn’t the outage. It’s not knowing what you missed.

Why we built HookDeploy the way we did

Every one of those cleanup sessions taught me the same lesson: the value isn’t in preventing the outage, you can’t, that’s not your infrastructure. The value is in never losing visibility into what happened, and having a real way to recover once things come back.

That’s the whole premise behind HookDeploy. A webhook hits us first, gets captured, gets stored, and only then gets forwarded on. If your downstream service is down, misconfigured, or mid-outage when it arrives, we still have it. When things come back, replaying it is a button, not a week of manual reconciliation. If you need that same event to go multiple directions, to a primary system and a backup system, or across regions, that’s a config, not a project.

Bad config downstream. No problem, replay it once it’s fixed. Provider outage. No problem, the events are sitting there waiting, not lost in the gap. Need to fan one event out to three destinations. No problem, that’s the default, not a workaround.

The cheapest insurance you’ll buy

I don’t say that as a marketing line. I say it because I’ve done the math on the other side of it, sitting in the aftermath of an outage, manually reconstructing what should have just been sitting in a log somewhere. What used to take days of reconciliation now takes seconds to replay. That gap, days versus seconds, is the entire product.

You don’t think about insurance until the day you need it. I’d rather you have it before that day comes.