Servers for Linux Release Notes

Servers for Linux Release Notes

2016, May 5 - 11:15 Download

Notes

Servers for Linux has been rebuilt against OpenSSL 1.0.1t to mitigate against CVE-2016-2108 and CVE-2016-2107.

2016, January 12 - 02:27 Download

New features

  • Added the -r command line option and corresponding host_root config file setting to set an alternate root for finding filesystems like /proc, /dev etc. This is primarily to support CoreOS so that the real /proc and /dev filesystems can be mounted inside the CoreOS container under a directory like /host. You would then set this option to /host and this will force LSM to look in /host/proc/whatever instead of directly looking for /proc/whatever.
  • No longer log proxy credentials in the log file on startup in order to improve log file security.
  • Prevent rapid respawning of LSM if there is an error on startup due to things like signals.
  • Cleaned up the logging of various messages related to Docker in order to reduce the "noise" in log files.
  • No longer report virtual ethernet interfaces on Linux that begin with the string ve-.
2015, September 22 - 10:34 Download

Notes

This is primarily a bug fix release although some new features have been added to support forthcoming New Relic services. These release notes will be updated with details soon.

Fixes

  • Disabling NFS no longer erroneously disables Docker, and the disable_docker config option now works correctly.
  • The ignore_reclaimable option is now enabled by default. This can affect your memory graphs at the point of upgrading.
  • Improved the detection of kernel processes.
  • No longer ignore processes with the / character in their name. This was resulting in a lot of processes not being reported, so it is to be expected that the process list will change after upgrading.
  • Recognize file-systems of type zfs on Linux as local filesystems.
2015, June 19 - 10:00 Download

Notes

To enable new Docker features, you need to grant the newrelic user access to the Docker socket, or provide alternative Docker connection information in nrsysmond.cfg.

New features

  • Docker containers now visible in application resources table

    Previously, individual Docker container information was not available on the APM Overview page; performance metrics were reported for servers only. The application resources table now displays separate performance metrics for active Docker containers, as well as relationships between containers and servers.

  • Docker metrics visible in Servers

    A new Docker-specific Servers page provides container-level metrics for the selected host, rolled up by image. The page includes CPU and memory usage, and the number of running containers for each image.

  • Show related Docker apps in Servers

    The Server Overview page includes a list of apps running in Docker containers, and provides detailed performance metrics for each container.

Bug fixes

  • SmartOS process CPU % normalized

    In SmartOS, a process can consume as much CPU as is available to its zone. Servers for Linux would faithfully report these values. Now the server monitor reports CPU as a percent of the zone's capacity. This means that values will be in the range of 0% to 100%, where 100% means full utilization of the zone's capacity.

  • Installation via RPM works for SELinux

    SELinux enforces constraints on users and their home directories. In the previous release of Servers for Linux, the newrelic user's home directory did not match this policy. This would cause the installation to fail. The home directory has been moved to /opt or /var/lib to satisfy SELinux.

  • Detect and skip stale NFS drives

    The agent pings an NFS drive before querying it for disk statistics. An issue existed where these pings would hang when pinging a stale drive, resulting in unusually long harvest cycles. The agent now circumvents the ping to avoid the issue.

  • Fixed double network values

    In rare circumstances the host operating system will return a network device twice. This caused the agent to record the network statistics twice for the device, resulting in the values being doubled. We now ignore duplicate network devices.

2015, March 4 - 13:57 Download

New Features

  • Ability to provide Docker container to host mappings

    This will enable upcoming features for Docker users in the New Relic UI. To enable these features, users will need to grant the newrelic user access to the docker socket, or provide alternative Docker connection information in nrsysmond.cfg.

Bug Fixes

  • Fix issues impacting SELinux installations.

    SELinux systems would fail to install Server Monitoring for Linux if the newrelic group did not already exist, and if the home directory did not satisfy Filesystem Hierarchy Standards. The Server Monitoring for Linux installation scripts have been improved to resolve these issues.

2015, January 19 - 16:00 Download

New Features

  • Ability to disable NFS Monitoring

    For those who do not want to monitor NFS mounted disks, the NFS monitoring feature can be disabled. Simply set disable_nfs=true in the configuration file.

Bug Fixes

  • Fix a race condition in upgrade script for RPM based installations which stopped the agent.

    After moving to the improved code base in v2, a race condition was exposed in some upgrades using RPM based installations. The race condition exists in the %preun scriplet which is executed when a package is replaced. Therefore the race condition exists in already installed agents and can occur when upgrading from any version prior to 2.0.2.111.

  • Properly create home directory in RPM based installations.

    When the newrelic user is created in new installations the /.newrelic home directory will also be created.

  • Better warning when experiencing permission issues.

    When monitoring file systems an error will occur if the newrelic user cannot access the directory. We've improved the messaging to help system administrators understand why their filesystem is not being reported.

2015, January 6 - 16:00 Download

Errata

This release is also 2.0.1.105

Bug Fixes

  • Fix an issue which prevented hostname overrides from propagating.

    The hostname for a Linux server can be overridden in the nrsysmond.cfg file. This change will make sure that setting is propagated to our servers.

  • Fix an issue on CentOS/RHEL systems which prevented the newrelic user's home directory from being created.

    Previously the home directory would be set to /.newrelic, however it was not being created during the RPM installation process.

2015, January 5 - 15:00 Download

New Features

  • Support for NFS v2/v3 mounted volumes

    Network attached storage disks mounted using NFS v2 or v3 will report for servers using Linux Server Monitor.

Bug Fixes

  • Remove a misleading warning message about configuration options greater than 1024 characters

    Configuration parsing has been updated and has much better support for large configuration options.

  • Agent keeps trying to connect after collector has been unavailable for a significant amount of time.

    During a prolonged network outage agents would give up trying to connect to New Relic. Now the agent is resilient to indefinite outages and will eventually re-establish the connection.

  • Fixes problem where installation did not properly set user newrelic's home directory

    The newrelic user had the incorrect home directory. Upgrades and fresh installations will fix the home directory to be /.newrelic

2014, November 12 - 17:05 Download

Bug Fixes

  • Fix an issue parsing multiple labels in the configuration for a single server

    Set one or more label names and values using the format category:value; for example, Environment:Production. To list multiple labels, separate each category:value with a semicolon (;); for example,Environment:Production;Environment:Dev;Environment:Staging.

  • In Joyent SmartOS zones we now calculate CPU utilization relative to the CPU Cap

    Previously CPU was calculated as the amount of user/system/io_wait etc. time that was spent on the CPU. In Joyent SmartOS zones, the virtual host can access many physical CPUs up to a cap set on the zone. We now report CPU utilization as a percentage of that cap instead of seconds spent per minute.

2014, October 16 - 16:30 Download

New Features

  • Set labels for your server directly in your configuration

    By setting the labels attribute, you can configure your server to register a set of labels. Labels can be used for filtering, roll-ups, and more to come.

Bug Fixes

  • Configuration file nrsysmond.cfg is checked for in /etc, /etc/newrelic/, and /usr/local/etc as well as the current directory

    Previously the configuration file has to be located in the agent's run directory or /etc/newrelic. This change aligns Servers for Linux with general standards.

2014, June 11 - 15:30

Notes

Bug Fixes

  • Fixed reporting of network metrics on servers with 'dummy' interfaces

    Some servers were observed to have many 'dummy' interfaces reported by the kernel. These interfaces could potentially use up the network metrics reported to New Relic. These 'dummy' interfaces are now ignored.

2014, June 5 - 16:00

Bug Fixes

  • Fixed reporting of network metrics on machines with many interfaces

    Machines which had more than 64 eligible network interfaces for metric reporting would not send any network interface metrics.

2014, May 5 - 22:09

Notes

This release changes the default communication behavior to use SSL by default. It also fixes several issues and increases robustness and supportability.

Upgrade Notices

  • All customers running on old x86 hardware that does not support the SSE3 instruction set (such as early releases of the AMD Opteron) should upgrade to this release as soon as possible. Prior to this release, the server monitor inadvertently contained SSE3 instructions which would cause an illegal instruction on such hardware. The only way the SSE3 instructions were executed was when we changed the choice and priority of SSL cipher algorithms at our data center, and we would like to change those priorities by the end of 2Q2014.

New Features

  • Support for custom SSL trust stores

    It is now possible to configure the server monitor to use a custom SSL trust store to authenticate New Relic's servers. In most cases this should not be necessary. The server monitor ships with the necessary certificates and will use them unless configured not to do so. Caution should be used with these settings. If the trust store does not contain the necessary certificates, the server monitor will be unable to report data to New Relic. See the documentation for more details.

  • Default to use SSL communication by default

  • Exclude virtual network interfaces by default

    Virtual network interfaces (e.g., "veth*" or 12-character hexadecimal names) could cause the server monitor to gather too many metrics. This would manifest itself as some data being missing in the New Relic charts and graphs for this server.

    To prevent this situation, the number of reported network interfaces is now limited to 64. These virtual network interfaces are also excluded from the metrics reporting.

  • Augmented verbosedebug logging for increased supportability:

    • Added logging of configuration settings
    • Added more environmental diagnostics to log on startup
    • Added logging of raw netio, diskio, disk usage and cpu values as reported by the Linux kernel
    • Added logging when metric limit (2000) exceeded
    • Added logging about the sender process of a SIGTERM signal
    • Increased logging around sleep and time functions

Bug Fixes

  • Fixed potential segmentation fault if user or group entry is not found

  • Made metric delta and accumulation functions more defensive against overflow

  • Only do extra environmental diagnostic logging when verbosedebug is on

  • Fixed nrsysmond command-line switch overrides

    Functionality to override configuration values via nrsysmond command-line switches was not working properly. This has been fixed.

2013, November 19 - 19:14

Notes:

  • Improved usefulness of "the fullest disk" reporting

    Excluded /boot from the filesystem reporting so that reporting "the fullest disk" is more useful. Previously, the inclusion of /boot partition could cause "fullest disk" alerting to frequently complain about small /boot partitions.

  • Improved startup robustness after reboot on systems which had tmpfs-like /var/run

    The nrsysmond would not reliably startup after a machine reboot due to incorrect assumptions (as observed on Ubuntu) about /var/run/newrelic still existing. On systems that have a temporary /var/run, you may have had to previously work around this issue by re-creating /var/run/newrelic in your puppet, chef, or custom init scripts. This has been fixed.

  • Improved robustness of disk monitoring

    Incorrect assumptions in the disk monitoring could cause invalid memory access (as observed on CloudLinux). This has been fixed.

  • Fix for over-reporting of memory and CPU usage of the server monitor process

    The server monitor process (i.e., nrsysmond) could over-report unexpected and incorrect CPU and memory usage for itself in the process list. This has been fixed.

  • Fix for some processes not reporting

    When sampling the running processes, groups with a large number of users in them could cause a failure to retrieve and report the process information. This has been fixed.

2012, August 23 - 18:31

Notes:

  • Fix a major process aggregation bug, where unrelated processes would be incorrectly aggregated together if a process exits between samples.
  • Numerous improvements and fixes for the Server Monitor Debian/Ubuntu and RHEL/CentOS packages.
  • Use CLOCK_MONOTONIC where available to work around quirks using system real time.
  • Report "nice" process time as "user" time.
  • Fix a minor timing bug in the disk I/O sampling code.