Fixes for host metrics and collector mode
Changed
-
Bundled certificates have been updated.
-
Use system-specific operation name in messaging span names.
When
messaging.operation.nameis present (e.g.Queue.add), it is used directly as the span name prefix:Queue.add (myQueue). When only the genericmessaging.operation.typeis available (e.g.send), the word "message" is added for clarity:send message (myQueue).
Fixed
-
Continue reporting non-disk host metrics when a mount is frozen.
-
Emit the
appsignal.action_nameattribute instead ofappsignal.root_namewhenset_root_nameis used in collector mode. The collector only reads the action name fromappsignal.action_name, so root names set this way were previously ignored, falling back to the OpenTelemetry span name. -
Update the OpenTelemetry span name directly when
set_nameis used in collector mode, instead of setting theappsignal.nameattribute. The collector uses the span name as-is, so names set this way were previously ignored. -
Emit the
db.system.nameanddb.query.textsemantic-convention attributes instead ofappsignal.sql_bodywhenset_sql_bodyis used in collector mode. This fixes SQL query sanitization when using the experimental collector mode. -
Fix host-metrics leaking zombie
[timeout]processes in Alpine linux containers.Before this release AppSignal agent relied on a proper init process that reaps child processes killed by system
timeout. Now the agent terminates and reaps unresponsive child processes in host-metrics collection and a subreaper is no longer required. -
Emit request parameters and headers using the attribute names the collector recognizes when in collector mode.
set_paramsnow emitsappsignal.request.payloadinstead ofappsignal.request.parameters, andset_headeruses thehttp.request.headerprefix instead ofappsignal.request.headers. The collector and server do not recognize the previous names, so this sample data was previously dropped in collector mode.
View the Python package v1.6.5 changelog for more information.