Countly NodeJS SDK 26.8.0

• ! Minor breaking change ! SDK behavior settings are now enabled by default. Changes made in SDK Manager on your server affect the SDK's behavior directly and take precedence over the matching init options:

  • The SDK fetches the settings at start, every 4 hours by default and after a device ID change, also before consent is given, and stores them for the next start

• ! Minor breaking change ! The SDK now stores its data in a .countly-storage folder in the app's working directory:

  • Before, it used a data folder inside the SDK package, which every reinstall or upgrade of the package deleted, so the next start counted as a new user
  • Data at the old location is moved to the new folder once, at the first start, and the old files are kept with a .migrated suffix. If the new folder is not writable, the old location stays in use
  • A relative storage_path is now resolved against the app's working directory, and the files go inside the given folder also when the path has no trailing slash
  • CountlyBulk without storage_path now stores its data in .countly-storage/bulk/<app_key>
  • Add .countly-storage/ to your .gitignore. Command line tools started from different folders and apps in containers should set an absolute storage_path to keep one device ID
  • The SDK prints a warning, also when logging is off, when the storage folder is not writable, for example on AWS Lambda, or is shared by two instances

• ! Minor breaking change ! The SDK now keeps the process running only while it has data to send and the server answers, so a script ends by itself once its data is delivered. Data that cannot be sent yet stays stored and is sent at the next start

• ! Minor breaking change ! With track_errors, unhandled promise rejections are now recorded as non-fatal errors, as the process keeps running after them. Uncaught exceptions are still recorded as fatal crashes

• ! Minor breaking change ! In a cluster, the workers and the main process now exchange messages that have a cly property. Your own message handlers receive them as well

• Added the init option standalone_worker that lets a cluster worker send its own data when the main process does not run the SDK, for example in PM2 cluster mode:

  • Each worker keeps its data in its own worker-<number> folder, numbered by PM2's instance number or else by the cluster worker ID
  • The cluster worker ID changes when a worker restarts, so pass device_id there to keep one device ID

• Added the init option behavior_settings for providing SDK behavior settings, in the format SDK Manager exports, until the first successful fetch

• Added the init option disable_sdk_behavior_settings_updates that stops fetching SDK behavior settings from the server. Stored and provided settings still apply

• Added SDK health check requests after init, also sent before consent is given. They can be turned off with the new init option disable_health_check

• Added an automatic backoff mechanism that by default pauses sending for 60 seconds after the server needed 10 seconds or more to answer. It can be turned off with the new init option disable_backoff_mechanism

• Added the init option request_timeout for setting, in seconds, how long a request waits for the server before it is sent again later (default 30, also for CountlyBulk)

• Added the method flush that sends all recorded data right away and resolves once it is delivered, for example before process.exit()

• A cluster worker whose main process does not run the SDK now prints a warning, also when logging is off, that its data is not sent

• CountlyBulk now prints a warning when salt is set

• Raw requests made with request now count towards the request queue size

• Queued requests are now sent one after another without waiting for the next heartbeat, also in CountlyBulk

• File storage now optimized for repeated writes

• The SDK now removes its process and cluster event listeners when halt is called

• Mitigated an issue where a missing storage file, which is normal at the first start, was logged as a warning

• Mitigated an issue where a request could be lost when the process ended while the request was being sent

• Mitigated an issue where calling process.exit() right after an SDK call could empty the storage file and lose the stored data and device ID

• Mitigated an issue where a BigInt value or a self-referencing object in recorded data crashed the app. BigInt values are now sent as text

• Mitigated an issue where a request that got no answer, or only part of one, stopped all further sending until the app restarted, also in CountlyBulk

• Mitigated an issue where CountlyBulk instances, and CountlyBulk next to the SDK, shared their storage, so data could be lost, sent twice or sent under the wrong app

• Mitigated an issue where removing consent was never sent to the server

• Mitigated an issue where a consent change made right before a device ID change, or before flush and process.exit(), was sent under the new device ID or lost

• Mitigated an issue where, after change_id without merge, the previous user's open view, unsaved custom properties, breadcrumbs and remote config values carried over to the wrong user

• Mitigated an issue where headers set in http_options made every GET request fail

• Mitigated an issue where internal limits were used wrongly

• Mitigated an issue where requests containing an apostrophe failed the checksum check of Countly 26.01 and newer servers when salt is set

• Mitigated an issue where a value containing half of an emoji could never be sent and stopped all further sending. Such characters are now sent as "�"

• Mitigated an issue where a CountlyBulk user created with ip_address threw an error on most calls

• Mitigated an issue where giving consent to a CountlyBulk user crashed the app when a call was waiting for that consent

• Mitigated an issue where the CountlyBulk start callback reported an empty queue while data was still unsent

• Mitigated an issue where calling track_errors repeatedly recorded each error several times

• Mitigated an issue where CountlyBulk replaced explicitly provided hour, dow and tz values of 0 with the current time in add_request and add_bulk_request

• Mitigated an issue where memory storage wrote files when a fatal crash was recorded