About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
Direct answer: Hosted line-of-business systems often call outward—HTTP webhooks to your ERP, SMTP email to staff, SMS gateways, payment status callbacks. Steady-state design means inventorying every destination URL, DNS name, TLS certificate, shared secret and firewall rule that must work after routine changes, not only after a migration. When callbacks cannot reach customer systems, symptoms look like “workflow stuck” while the application UI still loads.
Why it matters
Product and operations teams assume integrations are symmetrical: if users can reach the hosted app, the app can reach partners. In practice outbound paths use different DNS names, NAT rules, allow lists and secrets. A hosting move, certificate rotation, or office firewall change can block callbacks without affecting inbound remote desktop or browser access.
Webhook failures are often silent until queues fill: failed deliveries retry, then dead-letter, while business staff chase missing invoices or stock updates.
This article is steady-state webhook design. Post-move integration regression and inbound email or MFA cutover dependencies are separate planning topics.
How it works
Notification channels in typical hosted LOB estates:
| Channel | Direction | Common failure |
|---|---|---|
| HTTPS webhook | Hosted → customer URL | TLS name mismatch, 403 on new IP |
| SMTP relay | Hosted → mail relay | Auth change, SPF/DMARC rejection |
| Message bus bridge | Hosted → partner queue | Credential expiry, private network route |
| Platform push | Hosted → mobile vendor | API key rotation |
Webhooks are HTTP callbacks initiated by the sender (the hosted application). RFC 9110 defines HTTP methods, status codes and header field semantics; it does not define webhook-specific callback contracts—that is agreed per integration.
Security practice (practitioner guidance, with vendor example): prefer HTTPS, verify payloads with the sender’s documented scheme, and return precise status codes so senders backoff correctly. Stripe’s webhook documentation is one worked example: signature verification uses the raw body; timestamps and event IDs help reject replays and duplicates—other senders differ; follow their docs. Microsoft’s Graph change notifications overview illustrates subscription and callback URLs for cloud APIs; many LOB stacks are simpler or custom.
Design checklist (practitioner guidance—not a standard):
- Registered URL — production vs test; document redirect URI changes in change control.
- Authentication — HMAC header, mTLS, or bearer token; rotation plan with overlap.
- Source addressing — static egress IP versus dynamic; whether customer firewalls allow list hoster ranges.
- DNS and TLS — customer endpoint certificate SAN must match what the hosted stack validates.
- Idempotency — duplicate POST handling on the receiver; hosted side should log delivery IDs.
- Observability — retry policy, dead-letter queue, alert when failure rate exceeds threshold.
Hosted environments behind reverse proxies must preserve original client context only where needed; webhook outbound identity is usually the hoster’s egress, not the user’s browser.
What to choose and why
Push versus poll: webhooks suit near-real-time; scheduled poll suits partners who cannot expose HTTPS. Mixing both without documentation duplicates processing unless ids are shared.
Email as integration: acceptable for human notifications; risky as machine integration because spam filters and mailbox quotas break automation. Prefer structured webhooks or file drops for financial triggers.
Secrets management: store callback secrets in a vault or restricted config, not in clear text on multiple app servers.
Routine rotation (agreed overlap): for planned secret replacement, agree overlap with each partner—dual-trust periods are contractual and technical, not something to improvise on production endpoints.
Suspected or confirmed compromise: treat as an incident. Revoke and contain promptly under your incident procedure; do not delay revocation of a compromised secret until every partner has acknowledged rotation. Notify affected partners, rotate credentials, and review logs for abuse in parallel—routine overlap rules do not apply to active leakage.
Read next
Related articles you may want next.
- HostingChecking integrations still work after a hosting moveCheck business integrations after a hosting move. Test approved file transfers, notifications and connected workflows, with evidence of results and a plan for failures.Read article
- HostingKeeping email notifications and MFA working when the application server movesKeep email notifications and staff sign-in working through a hosting move. Check mail relays, registered application addresses and authentication dependencies with controlled tests.Read article
- HostingOutbound email as a hidden dependency of hosted business applicationsWhat breaks in hosted LOB workflows when SMTP relay, authentication or sender policy fails—even when interactive login still works.Read article
Risks and limits
This is steady-state webhook and notification design, not API gateway disaster recovery sequencing, not inbound email dependency articles, and not instructions to expose undocumented public APIs on a hoster platform.
No webhook endpoints were registered or fired in lab (untested). Third-party SaaS behaviours vary; confirm retry and signature headers in vendor docs at implementation time.
Sources
- RFC 9110 — HTTP Semantics — HTTP methods, status codes and headers.
- Microsoft Learn — Graph change notifications overview — subscription callback model (vendor example).
- Stripe — Webhooks — signature verification and replay considerations (vendor-specific example).
Learning notes
- Inbound user access working does not prove outbound webhooks work.
- Signature verification and HTTPS are baseline, not optional extras.
- Firewall allow lists must include hoster egress, not only office IPs.
- Retry storms need monitoring like application errors.
- Compromised secrets require prompt revocation—not partner-overlap delays.
This article was originally published on .