Upgrade to v4.x

Support for the following Kubernetes environment variables have been removed: ELASTIC_APM_KUBERNETES_NAMESPACE, ELASTIC_APM_KUBERNETES_NODE_NAME, ELASTIC_APM_KUBERNETES_POD_NAME, and ELASTIC_APM_KUBERNETES_POD_UID. The correct environment variables for these config vars are without the ELASTIC_APM_ prefix — for example KUBERNETES_POD_NAME — and has been documented that way since v2.11.0.

How to check. Search for any usage of ELASTIC_APM_KUBERNETES_ in your project. For example, using ripgrep, run rg ELASTIC_APM_KUBERNETES_. If there are any hits, remove the ELASTIC_APM_ prefix.

Support for filterHttpHeaders config option has been removed. Redaction of HTTP headers and also request cookies is controlled by the existing config option sanitizeFieldNames.

How to check. Search your project for rg filterHttpHeaders or rg ELASTIC_APM_FILTER_HTTP_HEADERS and remove them.

The default value of the useElasticTraceparentHeader config option has changed to false. This means that the vendor-specific elastic-apm-traceparent header will no longer be added to outgoing HTTP requests by default. The traceparent header (from the W3C trace-context standard) is added by the APM agent. If you need the agent to continue sending elastic-apm-traceparent HTTP header you can set it to true via env var or start options.

How to check. Search your project for rg useElasticTraceparentHeader or rg ELASTIC_APM_USE_ELASTIC_TRACEPARENT_HEADER.

The "patch" value for the contextManager config option has been removed. This was a limited async context management that predated the preferred AsyncLocalStorage core Node.js mechanism for context tracking. As well, the related and deprecated asyncHooks config option has been removed. Both were deprecated in v3.37.0.

How to check. Search your project for rg -w contextManager or rg -w ELASTIC_APM_CONTEXT_MANAGER which set the value to "patch". Also search for rg -w asyncHooks or rg -w ELASTIC_APM_ASYNC_HOOKS which set the value to false. If so, that config setting is no longer supported.

The logUncaughtExceptions config option has been removed. In v3 and earlier, when the APM agent was capturing an uncaught exception setting logUncaughtExceptions: true would tell the agent to print the error details to stderr before exiting; but logUncaughtExceptions was false by default. In v4, printing the error to stderr is done by default (to mimic the default Node.js uncaught exception behavior) and there is no option to disable that.

How to check. Search your project for rg -w logUncaughtExceptions or rg -w ELASTIC_APM_LOG_UNCAUGHT_EXCEPTIONS and remove any usages.

Support for the erroneous ELASTIC_SANITIZE_FIELD_NAMES and ELASTIC_IGNORE_MESSAGE_QUEUES config environment variables has been removed. The correct env vars are ELASTIC_APM_SANITIZE_FIELD_NAMES and ELASTIC_APM_IGNORE_MESSAGE_QUEUES, respectively, and were supported starting in v3.36.0.

The apm.startTransaction() method has been changed to return a do-nothing no-op Transaction, if the agent is not yet started. The return type has changed to no longer include | null. The intent of these changes is to allow the user to use .startTransaction() without having to worry if the agent is yet started, nor to have to handle a possible null return value.

How to check. Search your project for rg '\.startTransaction\b'. If your code handled a possible null return value from this function call, you can remove that handling.

The subtype and action properties have been removed from Transaction. This also impacts apm.startTransaction([name][, type][, options]) and transaction.setType(...), both of which now no longer accept subtype and action parameters. These two properties were deprecated in v3.25.0.

How to check. Search your project for rg '\.startTransaction\b'. If your code passed in subtype or action arguments, e.g. apm.startTransaction('a-name', 'a-type', 'a-subtype', 'an-action', { /* options */ }), then those need to be updated. Also search your project for rg '\.subtype\b' and rg '\.action\b'. If those property accesses are on an APM Transaction object, then you should remove them. (Note that subtype and action on APM Span objects remain in the API.)

The span.toString() and transaction.toString() methods have been removed as documented APIs. They were never in the "index.d.ts" types and were deprecated in v3.23.0.

Since v2.17.0 they would return a string of the form trace.id=<hex id the trace> span.id=<hex id of the span>, with the intent that this could be used in text-only loggers for log correlation. Using .toString() for this was deprecated in v3.23.0, and has now been removed in v4. In v4 the output of .toString() is not defined.

Instead, prefer the use of span.ids, transaction.ids, or apm.currentTraceIds. The v3 format may be reproduced via:

const {stringify} = require('querystring');
console.log(stringify(span.ids, ' ', '='));

For log correlation with structured logs, see Log correlation.

The apm.destroy() method is now async. Almost no users should need to use this method. However, if used, to be sure to wait for APM agent shutdown to be complete, one can now await apm.destroy().

{"log.level":"warn","@timestamp":"2023-08-04T16:54:03.116Z","log":{"logger":"elastic-apm-node"},"ecs":{"version":"1.6.0"},"message":"units missing in duration value \"5\" for \"metricsInterval\" config option: using default units \"s\""}

Configuration options that define a duration, like metricsInterval or exitSpanMinDuration, expect to have their units specified in the value (e.g. "10s", "100ms"). While current duration options have a default unit, to avoid ambiguity the APM agent will now warn if the units are not provided.

Byte size config options like apiRequestSize expect to have the size units specified in the value (e.g. "10kb", "1gb"). If the unit is not in the value, the agent will warn about it and fallback to bytes (b).