Endora

Features:

• Ability to upload logs to different Scalyr team accounts by specifying different API keys for different log files. See RELEASE_NOTES for more details.
• New configuration option default_workers_per_api_key which creates more than one session with the Scalyr servers to increase upload throughput. This may be set using the SCALYR_DEFAULT_WORKERS_PER_API_KEY environment variable.
• New configuration option use_multiprocess_copying_workers which uses separate processes for each upload session, thereby providing more CPU resources to the agent. This may be set using the SCALYR_USE_MULTIPROCESS_COPYING_WORKERS environment variable.
  Improvements:
• Linux system metrics monitor now ignores the following special mounts points by default: /sys/*, /dev*, /run*. If you want still capture df.* metrics for those mount points, please refer to RELEASE_NOTES.
• Update url_monitor so it sends correct User-Agent header which identifies requests are originating from the agent.

Misc:

• The default value for the k8s_cri_query_filesystem Kubernetes monitor config option (set via the SCALYR_K8S_CRI_QUERY_FILESYSTEM environment var) has changed to True. This means that by default when in CRI mode, the monitor will only query the filesystem for the list of active containers, rather than first querying the Kubelet API. If you wish to revert to the original default to prefer using the Kubelet API, set SCALYR_K8S_CRI_QUERY_FILESYSTEM the environment variable to "false" for the Scalyr Agent daemonset.
• New global_monitor_sample_interval_enable_jitter config option has been added which is enabled by default. When this option is enabled, random sleep between 2/10 and 8/10 of the configured monitor sample gather interval is used before gathering the sample for the first time. This ensures that sample gathering for all the monitors doesn't run at the same time. This comes in handy when running agent configured with many monitors on lower powered devices to spread the monitor sample gathering related load spike across a longer time frame.

Bug fixes:

• Fix to make sure we don't expect a valid Docker socket when running Kubernetes monitor in CRI mode. This fixes an issue preventing the K8s monitor from running in CRI mode if Docker is not available.
• Fix line grouping code and make sure we don't throw if line data contains bad or partial unicode escape sequence.
• Fix scalyr_agent/run_monitor.py script so it also works correctly out of the box when using source code installation.
• Update Windows System Metrics monitor to better handle a situation when disk io counters are not available.
• Docker monitor has been fixed that when running in "API mode" (docker_raw_logs: false) it also correctly ingests logs from container stderr. Previously only logs from stdout have been ingested.

Hydrus

Features:

• Add new initial_stopped_container_collection_window configuration option to the Kubernetes monitor, which can be configured by setting the SCALY_INITIAL_STOPPED_CONTAINER_COLLECTION_WINDOW environment variable. By default, the Scalyr Agent does not collect the logs from any pods stopped before the agent was started. To override this, set this parameter to the number of seconds the agent will look in the past (before it was started). It will collect logs for any pods that was started and stopped during this window. This can be useful in autoscaling environments to ensure all pod logs are captured since node creation, even if the Scalyr Agent daemonset starts just after other pods.

Improvements:

• Improve logging in the Kubernetes monitor.
• On agent start up we now also log the locale (language code and encoding) used by the agent process. This will make it easier to troubleshoot issues which are related to the agent process not using UTF-8 coding.
• Default value for tcp_buffer_size Syslog monitor config option has been increased from 2048 to 8192 bytes.
• New message_size_can_exceed_tcp_buffer config option has been added to Syslog monitor. When set to True, monitor will support messages which are larger than tcp_buffer_size bytes in size and tcp_buffer_size config option will tell how much bytes we try to read from the socket at once / in a single recv() call. For backward compatibility reasons, it defaults to False.

Bug fixes:

• Fix a bug / race-condition in Docker monitor which could cause, under some scenarios, when monitoring containers running on the same host, logs to stop being ingested after the container restart. There was a relatively short time window when this could happen and it was more likely to affect containers which take longer to stop / start.
• Update code for all the monitors to correctly use UTC timezone everywhere. Previously some of the code incorrectly used local server time instead of UTC. This means some of those monitors could exhibit incorrect / undefined behavior when running the agent on a server which has local time set to something else than UTC.
• Fix docker_raw_logs: false functionality in the Docker monitor which has been broken for a while now.
• Update Windows System Metrics monitor to better handle a situation when disk io counters are not available.

Celaeno

Bug fixes:

• Fix scalyr-agent-2 status command non-fatal error when running status command multiple times concurrently or in a short time frame.
• Fix scalyr-agent-status command to not log config override warning to stdout since it may interfere with consumers of the status command output.
• Fix merging of active-checkpoints.json and checkpoints.json checkpoint file data. Previously data from active checkpoints file was not correctly merged into full checkpoint data file which means that under some scenarios (e.g. agent crashed after active checkpoint file was written, but before full checkpoint file was written), data which was already sent to the server could be sent twice. Actual time window when this could happen was relatively small since full checkpoint data is written out every 60 seconds by default.
• Fix Postgres monitor error when specifying the Postgres database_port in the agent config.

Betelgeuze

• Upgrade psutil dependency which incorporates many critical fixes. As part of the change, Windows Server 2003/XP is no longer supported.
• Small fix for the pywin32 library which is used in the Windows version.

Aqua

Features:

• Add new win32_max_open_fds configuration option which allows user to overwrite maximum open file limit on Windows for the scalyr agent process.

Bug fixes:

• Fix bug in packaging which would cause agent to sometimes crash on Windows when using windows event log monitor.

Alcor

Bug fixes:

• Fix formatting of the "Health Check:" line in ``scalyr-agent-2 status -v` command output and make sure the value is left padded and consistent with other lines.
• Fix reporting of "Last successful communication with Scalyr" line value in the scalyr-agent-2 status -v command output if we never successfuly establish connection with the Scalyr API.
• Fix a regression in scalyr-agent-2-config --upgrade-windows functionality which would sometimes throw an exception, depending on the configuration values.

Security fixes and improvments:

• Fix a bug with the agent not correctly validating that the hostname which is stored inside the certificate returned by the server matches the one the agent is trying to connect to (scalyr_config option). This would open up a possibility for MITM attack in case the attacker was able to spoof or control the DNS.
• Fix a bug with the agent not correctly validating the server certificate and hostname when using scalyr-agent-2-config --upgrade-windows functionality under Python < 2.7.9. This would open up a possibility for MITM attack in case the attacker was able to spoof or control the DNS.
• When connecting to the Scalyr API, agent now explicitly requests TLS v1.2 and aborts connection if the server doesn't support it or tries to use an older version. Recently Scalyr API deprecated support for TLS v1.1 which allows us to implement this change which makes the agent more robust against potential downgrade attacks. Due to lack of required functionality in older Python versions, this is only true when running the agent under Python >= 2.7.9.
• When connecting to the Scalyr API, server now sends a SNI header which matches the host specified in the agent config. Due to lack of required functionality in older Python versions, this is only true when running the agent under Python >= 2.7.9.

Ursa

Bug fixes:

• Fixed a regression in Scalyr Windows Agent cmdlet script (ScalyrShell.cmd) which prevents the agent from starting.

Titan

Features:

• The status -v command now contains health check information, and will have a return code of 2 if the health check has failed. New optional flag for the status CLI command -H returns a short status with only health check info. A new configuration feature healthy_max_time_since_last_copy_attempt defines how many seconds is acceptable for the Agent to not attempt to send up logs before the health check should fail, defaulting to 60.0. For more information, please refer to the release notes document.
• Kubernetes yaml has been updated to include a liveliness check based on the new health check info, which will cause a pod restart if the agent is considered unhealthy.

Bug fixes:

• Fixed race condition in pipelined requests which could lead to duplicate log upload, especially for systems with a large number of inactive log files. Log files would be reuploaded from their start over short period of time (seconds to minutes). This bug is triggered when pipelining is enabled, either by explicitly setting the pipeline_threshold config option or by using a Scalyr Agent release >= 2.1.6 (pipelining was turned on by default in 2.1.6).
• Fixed the misconfiguration in Windows packager which causes some number of the monitors to not be included in Windows version. This generates import errors when attempting to use monitors like the syslog or shell monitor.

Misc:

• compression_level configuration option now defaults to 6 when using deflate compression_type (deflate is the default value for the compression_type configuration option). 6 offers the best trade off between compression ratio and CPU usage. For more information, please refer to the release notes document.

Serenity

Features:

• New configuration feature k8s_logs allows configuring of Kubernetes logs similarly to the logs configuration but matches based on Kubernetes pod, namespace, and container name. Please see the RELEASE_NOTES for more details.

Bug fixes:

• Fixed race condition that sometimes resulted in duplicated K8s logs being uploaded on agent restart or configuration update.

Misc:

• The Windows package is now built using pyInstaller instead of py2exe. As part of the change, we are no longer supporting 32-bit Windows systems. Nothing else should change due move to pyInstaller.

Rama

Features:

• New configuration option max_send_rate_enforcement allows setting a limit on the rate at which the Agent will upload log bytes to Scalyr. You may wish to set this if you are worried about bursts of log data from problematic files and want to avoid getting charged for these bursts.
• New default overrides for a number of configuration parameters that will result in a higher throughput for the Agent. If you were relying on the lower throughput as a makeshift rate limiter we recommend setting the new max_send_rate_enforcement configuration option to an acceptable rate or "legacy" to maintain the current behavior. See the RELEASE_NOTES for more details.

Minor updates:

• Default value for max_line_size has been raised to 49900. If you have this value in your configuration you may wish to not set it anymore to use the new default.

Bug fixes:

• Fixed Syslog monitor issue causing the monitor to write binary strings to the log file.