Logistics Software: Building for Interruption, Not Uptime

Build logistics software for interruption as a normal state, not an edge case. Scans, dock moves, and proofs of delivery should commit on the device, sync later without duplicates, and reconcile after connectivity returns. What counts is how much work continues when a cloud provider, carrier API, or cell signal disappears, not the uptime percentage on a status page.

Last updated November 2026

Why is uptime the wrong target for logistics software architecture?

Uptime is about your servers. Freight moves on trucks, docks, and scanners, often in places those servers cannot reach. A 99.9% SLA still leaves about 43 minutes of downtime a month, and it says nothing about the far end of the connection, where a driver sits in a dead zone or a dock loses Wi-Fi while every dashboard stays green.

When the center fails, damage shows up quickly. During the June 12, 2025 outage tied to Google Cloud and Cloudflare, FreightCaviar reported backend outages of 60 to 150 minutes on Google Cloud. Project44 modules for tracking, LTL rating, and dispatch failed, Truckstop’s mobile and location features were temporarily disabled, and authentication failures locked users out.

Warehouse numbers tell a similar story, with a caveat. A Synergy Logistics report released in February 2026 says 84% of warehouses had at least one significant disruption in 24 months and nearly half idled assets because of software or connectivity interruptions. FreightWaves’ coverage puts the cost at $5,000 to $100,000 per hour. Synergy sells warehouse management software, and neither its announcement nor that coverage states a sample size or method, so treat those figures as directional.

Most resilience writing for logistics stops at the data center: multi-region failover, standby databases, queues between services. That work is worth doing, and it protects the server. The harder questions are what a device does with no server at all, and how the data gets honest again afterward.

What interruptions does logistics software have to survive?

Five kinds of interruption, each failing in its own way. Multi-region failover helps with only the first.

Interruption Public example What must keep working Design response
Cloud or identity failure June 2025: logins failed across freight platforms Signed-in workers finish current tasks Session grace periods and cached task lists
SaaS vendor outage November 2024: ransomware at Blue Yonder disrupted Morrisons’ fresh and produce warehouse systems, and Starbucks moved to manual pay calculations, per CyberScoop Receiving, picking, and shipping Regular exports under your control, plus a rehearsed manual procedure
Company-wide compromise 2017: NotPetya halted Maersk operations at 76 port terminals, with an estimated hit of $200M to $300M, and staff fell back on WhatsApp on private phones, per The Register Gate decisions and staff communication Isolated backups, an out-of-band contact tree, and a rebuild runbook
Site or device connectivity loss Dead zones and dock Wi-Fi gaps that no cloud status page reports Scans, proof of delivery, and driver apps Local commit with queued sync
Carrier or partner API failure Rate, label, or tracking endpoints timing out Rate shopping, label printing, and status updates Cached contract rates, queued label requests, and a circuit breaker per partner

The last two rows can stop a dock while your platform still reports 100%.

How do you keep a warehouse or driver working offline?

Five rules carry most of the load.

  1. Commit on the device first. Write the scan or signature to durable local storage before the screen confirms it.
  2. Mint an ID when the action happens. AWS’s Builders’ Library article on idempotent APIs, by Malcolm Featonby, describes the model: each request carries a caller-provided identifier, repeats get an equivalent response, and the same identifier with different parameters gets a validation error. Apply that pattern to every scan so a retried proof of delivery can never become two deliveries.
  3. Write the state and the event in one step. Updating a database and then notifying a broker is a dual write, and either half can fail. AWS’s transactional outbox guidance puts the event in the same transaction as the business data, then relays it. That relay can send duplicates, which is why rule 2 comes first.
  4. Keep two timestamps on every event: when it happened and when the system learned of it. A delivery signed at 2:10 and synced at 4:40 is a 2:10 delivery for detention and billing, and a 4:40 fact for anything asking what you knew at the time.
  5. Set authority before the outage. Two offline devices cannot both promise the last unit. Allocate stock, dock slots, or load assignments to a site or device in advance so each acts only on what it holds. Anything outside that allocation waits, and the app says so plainly.

What happens when the connection comes back?

Reconnect is the expensive part, and most guides we read give it only a sentence or two. Recovery has an order.

  1. Replay slowly. Devices that reconnect all at once create a load spike. Throttle the backlog and keep live traffic ahead of it.
  2. Apply updates by event time, per entity. A late “picked” event must not overwrite a newer “shipped” state.
  3. Quarantine conflicts. When a count disagrees, or a load was both reassigned and delivered, send it to a human queue with both versions attached. Last-write-wins has no business deciding physical inventory.
  4. Reconcile across systems. Line up device totals, WMS, TMS, carrier status, and invoices, then list every mismatch with an owner.

Close the incident when every mismatch is resolved or owned. Green service health is only halfway there. Keep the original events untouched while you correct them. Our piece on audit trails in regulated products covers that record.

Which legal clocks keep running during an outage?

Regulators do not pause for your vendor’s incident, and most resilience guides barely mention them. Two US examples show the design impact.

Rule What it requires Architecture consequence
49 CFR 395.34, ELD malfunction The driver gives written notice to the carrier within 24 hours, reconstructs the current 24-hour period and previous 7 consecutive days unless those records are already possessed or retrievable from the ELD, and keeps manual logs until the ELD is serviced. The carrier corrects the malfunction within 8 days of discovery or the driver’s notification, whichever occurs first, or requests an extension from FMCSA within 5 days of the driver’s notice. Keep those 8 days of duty records retrievable offline, and track the repair deadline as a dated ticket
FDA Food Traceability Rule (FSMA 204) Anyone who manufactures, processes, packs, or holds listed foods must give FDA records within 24 hours of a request, or within a reasonable time FDA agrees to Traceability data needs a retrieval path that does not depend on the primary WMS

On timing, FDA’s rule page says the agency proposed a July 20, 2028 compliance date and that Congress directed it not to enforce the rule before then. FDA’s Q&A also exempts transporters of food while treating distribution centers as covered, so a 3PL’s fleet and its warehouse can land on different sides of the rule.

The same discipline shows up in our HIPAA-compliant software development guide: decide what the record must prove, then design so an outage cannot break it. We aren’t lawyers, so confirm with counsel how each rule applies to your operation.

How do you test logistics software architecture for interruption?

Schedule scored drills. Run them in staging, on real devices.

  1. Kill single sign-on in the middle of a shift. Can signed-in workers still finish their tasks?
  2. Put a handheld in airplane mode for an entire shift, restore it, and compare every event.
  3. Block a single carrier API for an hour. Do labels queue and then release cleanly?
  4. Replay a day’s events three times, then out of order. Totals should not move.
  5. Cut the WMS and operate the site from the last export and the written procedure.
  6. Answer a 24-hour traceability or duty-record request while the primary system is down.

Capture three numbers: the share of work that continued, hours to zero open mismatches, and duplicates created. We found no public benchmark for these, so set targets from your own shipment cutoffs.

What should you check if AI tools wrote the code?

Start with the failure paths. A demo on office Wi-Fi never exercises a dead zone, a duplicate scan, or an out-of-order event, so those gaps can pass review, which is the pattern we describe in why AI writes code that passes review and fails at scale.

Search the codebase for confirmation screens that wait on an API response, retries with no idempotency key, a database write followed by a separate publish, and timestamps set by the server on receipt.

Our $4,950 Rescue Audit returns risk-ranked findings and a prioritized roadmap in 10 business days, credited toward remediation. When the fix needs sustained work, Delivery Pods start from $15K per month with 30-day notice.

Frequently asked questions

What is offline-first logistics software?

Offline-first logistics software records every action on the device first. Scans and signatures save locally and sync later, with unique IDs so retries cannot create duplicates.

What uptime should logistics software target?

Pick an uptime target, then add one for work that continues during an outage. A 99.9% target allows about 43 minutes a month and 99.99% about 4, but neither measures sites or devices that lose their connection.

How long should a warehouse run without its WMS?

No public benchmark exists. Work back from your tightest shipment cutoffs. If the site cannot pick and ship to the next cutoff from an export and a written procedure, that gap is your real exposure.

Book a Code Review

Unsure what your app does when the signal drops? Book a Code Review and our senior engineers, who average 10+ years of experience, will show you where it assumes a connection it will not always have.