Skip to content

Mobile apps

Observability → Mobile apps shows your apps’ crashes and performance from Firebase’s BigQuery export, read in your own Google Cloud project. Nothing is copied to Seylo. To set it up, see connect Firebase.

Pick an app, then a window of 1, 7 or 30 days. Each app is one platform of one Firebase app. The page says first what the numbers reach: which export it read and how far.

Every view names its export’s cadence and the newest event it read, because Google states a cadence, not a delay figure:

  • With streaming on and read, new crashes appear as Crashlytics streams them, and the daily batch table fills in behind it.
  • On the daily batch export alone, a crash can take a day or more to appear.
  • Performance Monitoring is a daily export.

A crash that happened a few minutes ago may not be on the page yet. Nightshift says the same: it tells you when an app’s data does not reach the alert’s window yet, and never claims a spike it cannot see.

Part What it shows
Crash-free users The share of installations that recorded a session and did not crash: one minus crashed installations over installations with a session
Issues Up to 50 groups of crashes, with their event counts and how many installations each hit
Filters App version, OS, device and error type. They are sent as parameters, never put into the query text

Crash-free users can be unavailable, and then the page says why, in words, instead of a number. The Crashlytics table holds crash events only, so the figure needs the sessions export. It is withheld when sessions are off, when the count could not be read, when no installation recorded a session in the window, and when the two exports disagree. See troubleshooting.

Select an issue to see its newest 5 events: the stack with the blamed frame marked, the exception type and message, the app version, OS version and device, its process state, and its custom keys, logs and breadcrumbs. Stack frames keep their file and line.

  • Email addresses, phone numbers, card numbers, tokens and keys are removed from messages, custom keys, logs and breadcrumbs. A name typed into free text is not recognised.
  • A user id is shown only when an admin turned on Show user ids on the connection. Otherwise it is never read.

Performance Monitoring’s export gives, for the window:

Part What it shows
App start How long the app took to start
Screens Up to 20 screens
Network requests Up to 30 URL patterns, each with its error rate: requests answered with a 4xx or 5xx over requests that got an answer

An error rate is not shown (and never shown as 0%) when no request in the group got an answer to count.

Each page view is a BigQuery query in your project, capped at 10 GiB billed. The page says what the read cost, and says so when it cannot be priced. See what each read costs you.

When a backend alert fires in a project with a connected app, Nightshift can read the app’s crashes as one lead among its evidence: the crash and installation rates and the blamed frame. A crash is a lead, not a cause, and Nightshift gives no verdict that a spike happened. It reads at most twice per run and only for apps the connection lists.

Everyone whose access is not limited to some namespaces can read it, because an app belongs to the project and not to a namespace. Only an admin changes the connection.