1.2.3

• Build gProfiler as an Aarch64 executable

1.2.2

• Python: upgrade py-spy to v0.3.10 and PyPerf to v1.1.2
• Java: upgrade to async-profiler v2.5 and allow using CPU mode with --java-async-profiler-mode=cpu.
• Disable PHP profiling by default (pass --php-mode phpspy to enable).
• Write metadata & metrics to local output files (in --output-dir mode).
• Limit the remote logs buffer size (avoid buffer growth if logs aren't being uploaded).
• Ruby & Python: improve removal handling of temporary files.

1.2.1

• Upgrade to py-spy 0.3.9
• Get process comm from /proc/PID/status instead of /proc/PID/comm (as the latter is missing in old kernels)
• python: py-spy: Use SIGKILL to stop py-spy when hanging

1.2.0

• Initial Aarch64 support (as a Docker container)
• Exit with 0 if gProfiler's mutex is already taken.
• Disable profilers whose start() method fails (instead of exiting gProfiler)
• Use /proc/pid/comm as the first stack of process profiles
• java: Allow skipping the JDK version check
• php: Log stderr & stdout if startup fails
• python: PyPerf: Allow passing 0 for --pyperf-user-stacks-pages

1.1.9

• Upgrade PyPerf to v1.1.1 to fix a memory leak
• Add memory usage logging to gProfiler's container (to be able to track gProfiler's total memory usage)

1.1.8

• Add metrics & metadata collection: gProfiler collects system metrics (CPU & memory utilization) & static metadata (OS version, hardware info, cloud metadata). These will augment the data displayed in the backend.
• java: Identify Java process to profile by existence of libjvm.so in their /proc/pid/maps.
• java: Allow sending detailed native samples in case there are no debug symbols; the extra data allows to add proper symbols information in the backend. Currently disabled, until backend support will be added.
• python: Add timeout to py-spy runs
• Bugfix: attempt to fix-up numerous output files situation of PyPerf (delete all & use the latest)
• Print nicer exceptions: if Python processes go down before profiling starts, if ldd is missing, if gProfiler failed to take its lock due to an error.

1.1.7

• Run perf continuously - instead of re-starting perf each profiling session, we now run it continuously and request it to dump data each session (improves accuracy & performance)
• Preliminary NodeJS support - this requires running your Node app with some extra flags, and to add --nodejs-mode perf to gProfiler's arguments
• Native stacks for Python (PyPerf based) - PyPerf now collects native stacks (CPython runtime, library DSOs, ...) in addition to the Python stacks & kernel stacks

1.1.6

• java: Allow different gProfiler versions to use different APs (#137)
• java: Build async-profiler with -static-libstdc++ (#139)

1.1.5

python, ruby: Remove collapsed files after reading (#128)

* merge: Add parse_and_remove_one_collapsed()

* python: Use parse_and_remove_one_collapsed()

Fixes a regression: we didn't remove old collapsed files.

* ruby: Use parse_and_remove_one_collapsed()

Fixes same bug from Python.

1.1.4

Server Logging (#119)

* Implement server logging

* Use new logger adapters across project

* Add server logging section to README.md

* logging: allow not sending extra to server

* Server Logging: Change environment variable name

* ruby: use logger adapter

* gprofiler: bump version