Blog / Build in public ·
What we changed after reading a week of subscription SDK bug reports
Every week I read public bug reports about RevenueCat, Superwall, Adapty, and Purchasely. This is what we changed in Payblast, and why.
I started Payblast as a RevenueCat alternative for my own apps. It is still early. Every week I read public complaints about RevenueCat, Superwall, Adapty, and Purchasely: GitHub issues, posts on X, community threads. The same classes of problem keep showing up. When they do, I try the same case against Payblast and change it if our answer is not good enough.
This week that meant three things: rotating a secret API key without taking entitlement checks down, noticing trial bursts and empty offerings before they sit in production, and making the React Native SDK fail closed by default.
All three of these changes are merged.
Rotate a secret without downtime
Last week Superwall revoked every organization's API keys after they found suspicious activity on their network. Server-side entitlement checks failed until people rotated. The customer email arrived a day or two later. Superwall said the SDK and paywalls kept working, and that only backends using an Organization API key were affected. The rotation itself is the right response to a breach. The painful part is the gap: keys die, backends start returning errors, and the email is late.
Payblast now lets you mint a new
sk_
key while the old one stays valid for 24 hours or 7 days. The public SDK key
(pk_) does not change, so you
do not ship an app update. Last used is recorded per key, at most once a minute, so you
can wait until the old row is quiet before the window ends. A scheduled job expires the
retiring key when the window is over. A sweep every 15 minutes catches anything that was
missed. If a secret leaked, Revoke stops that exact key immediately. Other keys on the
project stay valid.
Every change emails the project owner and, if a webhook URL is saved, posts api_key_created, api_key_expiring, api_key_expired, or api_key_revoked. The payload
has the prefix, never the full secret.
This sits next to store-credential health alerts that already shipped. If App Store, Play,
or Stripe credentials are rejected, Payblast emails you and posts store_credential_invalid.
When they work again, you get store_credential_recovered.
Those alerts do not wait a day or two.
-
A new
sk_key can overlap the old one for 24 hours or 7 days, or replace it immediately. - Last used updates at most once a minute on the exact secret that authenticated.
- Revoke is immediate and only for that secret.
- The owner is emailed, and the webhook fires, on create, expiring, expired, and revoked.
Trial bursts, empty offerings, Apple Ads IDs
A RevenueCat customer described new trials starting in bunches: 8 to 16 at once, same IP, different anonymous IDs, never converting. Staff suggested blocking the paywall by IP or using DeviceCheck. The original poster disabled trials for now.
Store notifications do not include an IP. On public SDK calls, Payblast stores an
HMAC-SHA256 of the client IP and its /24 or /48 network, keyed with the endpoint secret
and the project id.
Raw addresses are never stored. When a new trial is applied and 8 or more distinct
customers share that hash in 60 minutes, those customers and events are flagged suspected_trial_farming. The
owner is emailed, at most once per cluster per day. A webhook
trial_farming_suspected
goes out with a count and a fingerprint, not a customer. The dashboard shows a banner and
a badge. Purchases are never blocked. You can exclude flagged trials from the Trials KPI
if you want. The default is to include them.
A developer also posted that a current offering can ship with no products for one of the stores in a project, so that store's paywall comes back empty. Another asked for Apple Ads campaign, ad group, and keyword IDs next to the names. Names change. IDs do not.
-
The dashboard banner and
GET /v1/health(secret key) warn when the current offering has no products for an app's store, when the project has offerings but none is current, or when an app that had at least 6 trial or purchase events in the previous 28 days has had none in the last 7. Sandbox apps and apps with no history are ignored. -
GET /v1/offeringsincludes awarningsarray in production so a client can log an empty current offering. -
POST /v1/customers/:app_user_id/attributes(public key) accepts Apple Ads campaign, ad group, and keyword IDs and names. The same object is on customer info, event history, and customer webhooks. The customer page shows IDs next to names.
React Native SDK: fail closed, stay off the main thread, refetch
These changes are merged in the public React Native SDK.
RevenueCat's React Native wrapper defaults entitlementVerificationMode to
DISABLED and always sends that value,
which turns off native Trusted Entitlements even though the native SDKs default to INFORMATIONAL.
Payblast defaults to ENFORCED.
Purchases.configure({ apiKey })
does not send a verification mode unless the app sets one, so a missing JS option cannot
override the native default. Unverified StoreKit transactions are rejected. Play Billing
still only accepts PURCHASED.
A RevenueCat iOS issue reports getCustomerInfo() decoding cached JSON on the
main thread, hanging the app. A Purchasely question reports a hidden InitializerProvider that starts the SDK on
the main thread at process start, including background push wakeups.
Payblast does not keep a native customer-info cache, so there is no JSON decode of cached
customer info on the main thread. The iOS module already returned requiresMainQueueSetup() == false.
Purchases, restores, and StoreKit listeners now run on Task.detached. Android has no
ContentProvider
and no App Startup initializer. BillingClient
is created on a background executor on the first
purchase
or restore, not in the native-module constructor.
Only launchBillingFlow
hops to the UI thread.
If Superwall's iOS config fetch fails while the device is offline, configurationStatus stays failed.
getPaywall
throws "No Config" until the process restarts.
Payblast getOfferings()
always hits the network. A failed network fetch retries in the background with exponential
backoff (1s, 2s, 4s, capped at 60s), when the app returns to the foreground, and on the
web online
event. Concurrent requests share one in-flight call. 4xx and configuration errors are not
retried. Automatic retries wait at least 2 seconds so foreground and online events cannot
spam.
-
Entitlement verification is
ENFORCEDby default and is never silently overridden from JS. -
Purchase and billing work run off the main thread.
BillingClientis created lazily. - Offerings refetch with backoff and on foreground or online after an offline start.
If a subscription SDK has caused you trouble this week, I would like to hear what happened. Reply wherever you found this post, or email hello@bitscorp.co.
Try it with your own app
Create a project, attach your store products, and check one entitlement. One or two apps are free to start, with no card required. Send us the pain points you are still living with.