[ CONNECTOR RESCUE ]
Your connector is dying. Your order flow doesn't have to.
Middleware goes zombie all the time: acquired, sunset, abandoned, or just quietly unmaintained. Replacing it is routine work when it's done with a parallel run instead of a leap.
[ THE SYMPTOMS ]
You usually know before it's official
- ▸The vendor was acquired and the product went quiet: no releases, no roadmap, support tickets into the void.
- ▸The connector is "closed to new customers," which is vendor for end-of-life.
- ▸Every platform update breaks something, and the fixes now come from your side, not theirs.
- ▸Nobody at your company knows exactly what it syncs, when, or what happens when it fails.
- ▸The renewal invoice still arrives on time.
That last one has a number on it.For one distributor the renewal was $25,000 a year, for a connector nobody could fully explain. They own their integration outright now, and that line is off the budget for good.
[ THE RESCUE ]
Replace the wire while the current still flows
01
Capture the old connector's behavior
Before touching anything, we record what actually crosses the wire today: every payload, every field, every quirk the business has come to depend on without knowing it.
02
Build the replacement against reality
The new integration reproduces the old one's contract byte for byte where it's right, and fixes it deliberately where it's wrong. Every difference is a decision, not an accident.
03
Parallel run, then a config flip
Old and new run side by side on live data until the diffs read zero. Cutover is a URL change with a rollback path, done on a quiet morning, not a weekend of held breath.
ERPs we've done this against: Sage 100 in depth, Prelude and friends by the same method. If your system of record has an API, a file drop, or even a database we can read, it's rescuable. And the replacement isn't another rented black box: it's the owned core described on the Sage 100 ↔ Shopify page, so the rescue is also the last connector bill you pay.
[ AFTER THE RESCUE ]
What stable looks like
Recreated screens from a rescue we ran to completion, with identifying data replaced.
The parallel-run scorecard
Step 03 isn't a promise, it's a table: the replacement answering the same queries as the legacy path, verified byte for byte, at a fraction of the read time.
| Endpoint | Bridge | Live eLink (same query) | Fidelity |
|---|---|---|---|
| productlines | 0.9s | N/A | byte-identical |
| warehouses | 0.9s | N/A | byte-identical |
| products | 2.0s | N/A | prefix-identical |
| customers | 1.2s | N/A | prefix-identical |
| salesorders | 0.9s | 18.6s | prefix-identical |
| invoices | 0.9s | 9.2s | prefix-identical |
| salesorders/history | 7.3s (81,000+ rows) | 15.4s | prefix-identical |
The connector watches its own dependencies
When the ERP goes down, you hear it from the integration, with an all-clear when it recovers. Not from a customer, three days later.
🚨 webstore: Can’t reach Sage (eLink)
eLink server error (HTTP 500)
✅ webstore: Can’t reach Sage (eLink), recovered (was down ~6m)
Rescued doesn't mean unattended
The box audits itself: listening ports, logins, file changes. Anything new gets reported, even when the verdict is "harmless, here's why."
⚠️ [FORGESYNC-WARN] 0 critical / 1 warning finding(s)
WARNING (1)
- Unexpected loopback-only listening port(s): [3100]
Expected: [22, 80, 443, 3306, 6379]Currently listening: [22, 80, 443, 3100, 3306, 6379]Bound to 127.0.0.1/::1 only. Not reachable off-box.If legitimate, add to EXPECTED_LISTEN_PORTS in /opt/forgesync-ops/security-monitor/monitor.py.
Full details in the email to ops@acme-fab.com.
ip-10-0-2-114 · 2026-07-22 18:45 UTCOn a retired stack?If your connector era ended with Sage eBusiness Manager, we wrote a plain accounting of the four realistic paths forward and the questions to ask any replacement, including us:Sage retired eBusiness Manager. Here are your actual options.
How long does your connector have?
Bring what you know (or admit what you don't) about the current integration. The first deliverable is a map of what it really does, which is worth having even if you never replace it.
Start with a rescue assessment