Testing a Travel App Across Wi-Fi, Roaming and eSIM Connections
Travel apps fail not only when a phone has no signal, but also when the device looks connected, when a request hangs long enough for a second tap, and when the network path changes mid-write. This is a testing guide for
Travel apps fail not only when a phone has no signal, but also when the device looks connected, when a request hangs long enough for a second tap, and when the network path changes mid-write. This is a testing guide for those cases. It is not a trip report, and it does not claim the checks below were run as a field study.
What an eSIM changes — and what it does not
The GSMA describes eSIM as a new way to load a SIM into a device: the role of the SIM itself is not changing, and operator credentials still sit in a secure element so the device can authenticate onto mobile networks. For consumers, more than one operator profile can be stored and switched remotely.
That is the right mental model for tests. Enabling or switching an eSIM profile is SIM provisioning — installing or selecting subscription credentials — not, by itself, a distinct application-level IP networking stack. Once a profile is active, the app still sees some mix of Wi-Fi and cellular IP connectivity, with ordinary DNS, TLS, timeout and mid-transfer failures. Do not assume that "the user is on eSIM" implies a special routing API, a guaranteed path, or better performance than any other cellular subscription.
Treat a profile switch as a change of the active mobile subscription, then apply the same reachability, timeout and retry checks you would after any other interface change.
Offline state is not service reachability
Browsers expose navigator.onLine and fire online / offline events on window. These are hints, not a probe of your API.
MDN states that navigator.onLine is inherently unreliable. A LAN connection is generally treated as online even when that LAN has no internet access. On Windows, status may depend on reaching a Microsoft home server that a firewall or VPN can block, even when other internet access still works. MDN's guidance is not to disable features based on this flag — only to hint that the user may seem offline.
The online event has the same limit. MDN notes it should not be used to decide whether a particular website is available: network problems or firewalls can still block that site after the browser is marked online.
Captive portal Wi-Fi is a common instance of this gap. The device may appear online because it joined a local network, while app requests time out, fail TLS, or are redirected to a login page. Probe the service you actually depend on, and record the failure class (no route, DNS, TLS, timeout, unexpected redirect, 5xx) instead of collapsing everything into one offline boolean.
Timeouts, slow requests and path changes
On travel networks, slow and dead can look alike until you set a timeout. A request that waits indefinitely leaves the UI stuck and invites another tap.
AbortSignal.timeout() returns a signal that aborts after a specified active time with a TimeoutError DOMException. MDN notes that this timeout follows active rather than elapsed time — it is effectively paused in a suspended worker or in the back-forward cache — and that fetch can still fail for other reasons, including user abort. If you use the API, distinguish timeout, user or navigation abort, and other network or HTTP errors. Treat the timeout budget as a test input and log the failure class; do not present it as a measured result.
Also test path changes while a request is in flight: Wi-Fi to cellular, cellular to Wi-Fi, airplane mode, or an eSIM profile switch. In-flight work may hang or fail, and DNS or sockets may still point at the old path. The app should fail visibly and recover, not keep a spinner over a dead request.
Retries should not duplicate writes
A timeout often means the client does not know whether the server applied a write. Retrying a booking, payment or message send can create a duplicate if the first attempt already succeeded and only the response was lost.
Put that in the test plan, not only retries of reads. For operations that must not happen twice, the API needs a way to recognise the same intent on retry — for example a client-generated idempotency key, or a status check on an existing resource — if the backend actually supports that. If it does not, do not invent safety in the client. Test the unguarded double-submit, and prefer a status check or a disabled submit control over a second silent write when the outcome is unknown.
A practical test matrix
Use this as a lab checklist, not as evidence that every cell was executed here:
| Condition | What to look at |
|---|---|
| Trusted Wi-Fi, API reachable | Baseline success; onLine is only a hint |
| Cellular only (home or roaming) | Same probes; no special "eSIM stack" assumptions |
| Captive portal Wi-Fi |
onLine may be true; API probe fails or redirects |
| LAN up, API host unreachable | Online flag versus real failure class |
| Switch Wi-Fi ↔ cellular mid-request | In-flight failure, retry policy, UI recovery |
| Enable or switch an eSIM profile, then re-probe | Subscription change, then the same reachability checks |
| Airplane mode on, then off | Offline hint versus queued work and stale requests |
| Slow link or forced timeout | Timeout error versus a hang |
| Timeout on a write, then retry | Duplicate created or not; status-check path |
| Double tap on a submit control | Second request blocked or deduplicated |
Automate what you can against a staged API. Captive portals and operator switching often still need a manual pass on real devices. Report only the cases you actually ran.
Sources
- Navigator: onLine property — MDN
- Window: online event — MDN
- AbortSignal.timeout() — MDN
- GSMA: eSIM
Disclosure: I work on uPhone eSIM, a travel eSIM marketplace. This is a software-testing guide, not a product test or a comparison of providers. https://uphone.com
Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.