Overview - Ghost

Ghost

7 days30 days

Latest CI Pipeline Executions

Fuzzy

Succeeded
29510
a2b57f0a Made the members import inline threshold configurable The row count above which a member import is handed to a background job instead of being performed while the request is open was a literal 500 in the middle of the routing decision. How long a request may be held open before an import goes async depends on the deployment, so this is an operator concern rather than a constant that suits every host. It is now members.importer.inlineThreshold, defaulting to the value it replaces. The importer does not read config itself: it is constructed with a getInlineThreshold getter alongside the getTimezone and repository getters it already takes, and the composition root in service.js resolves the value, matching how VerificationTrigger is given its thresholds. A getter rather than a value because an operator can change the setting between requests. Config arrives untyped from defaults.json, an operator's file or an environment variable, so a small resolver at the boundary turns it into a number, accepting numeric strings and falling back to the shipped default when a higher layer supplies something unusable, so a bad override is correctable live rather than taking member import down. Tests drive the inline and deferred paths through the API with the threshold set per test, so a single small fixture stands on either side of it rather than the boundary needing a fixture large enough to cross the shipped default.
by Rob Lester
Succeeded
29510
19cecda6 🐛 Fixed members import failing on a file with no rows An empty upload — a file with no bytes, or headers with no rows — used to fail with a 500, "Cannot convert undefined or null to object": the importer derived its column list from the first parsed row and there was none. Nothing guarded this on the server. The upload middleware checks only that a file was attached and that its type is allowed, and the admin import modal catches the empty case in the browser, so it was reachable by anything driving the API directly. The importer now chooses its path explicitly rather than by accident. Nothing to import returns immediately, reporting that nothing was imported. An import small enough to finish while the request is open is performed there. Everything else is handed to a background job. An empty upload short-circuits in prepare() before any file is written or import is performed, so process() returns on the zero row count before it would reach for a file that does not exist — the branch is load-bearing rather than a fall-through that happened to produce the right result. The middle path reads as a question about cost rather than size: an import carrying populated Stripe data is deferred whatever its row count, because resolving Stripe data per row is slow. The count of rows to import was previously called batches, a name inherited from a batch size that had been fixed at 1, so it had equalled the row count for as long as the constant existed; it is now rowCount. Tests drive all three paths through the API, with the inline threshold set per test so the same small fixture stands on either side of it rather than needing a fixture large enough to cross the shipped default.
by Rob Lester