Incorrect handling of overcommit_memory kernel sysctl.

[Chris Friesen]

Title
-----
Incorrect handling of overcommit_memory kernel sysctl.

Brief Description
-----------------
In the code for collectd (and similar code for rmon) there is logic that reads /proc/sys/vm/overcommit_memory and treats it as 'strict" if the value is 1.

Unfortunately, this is wrong. A value of 0 means "heuristic overcommit", a value of 1 means "always overcommit", while a value of 2 means "don't overcommit". (See https://www.kernel.org/doc/Documentation/vm/overcommit-accounting)

The end result is that we end up raising critical memory alarms when there is still 50% of the memory unused.

Severity
--------
Major: it results in a spurious critical memory alarm, which causes the node to go "degraded"

Steps to Reproduce
------------------
Install StarlingX, configure with "--kubernetes", install openstack application. Active controller node will be degraded due to spurious critical memory alarm.

Expected Behavior
------------------
There is over 50% of memory still unused, should not have critical memory alarm.

Actual Behavior
----------------
There's a spurious critical memory alarm.

Reproducibility
---------------
Reproducible

System Configuration
--------------------
Dedicated storage, kubernetes

Branch/Pull Time/Commit
-----------------------
starlingx master branch, 2018-12-10_20-18-00

Timestamp/Logs
--------------
N/A

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/624821
Committed: https://git.openstack.org/cgit/openstack/stx-integ/commit/?id=0ec172537192932c11f7a9cdc799fbc7e49a22e1
Submitter: Zuul
Branch: master

commit 0ec172537192932c11f7a9cdc799fbc7e49a22e1
Author: Eric MacDonald <email address hidden>
Date: Wed Dec 12 17:15:10 2018 -0500

    Fix collectd Memory plugin Strict Mode learning

    Existing code sets overcommit strict mode to True
    if any non-zero value is returned from a read
    of /proc/sys/vm/overcommit_memory.

    This is incorrect.

    Strict mode should only be set when the returned
    value is 2.

    Change-Id: I2c5328624571bb3b2f478d5a79615650bb92cbd2
    Closes-Bug: 1808225
    Signed-off-by: Eric MacDonald <email address hidden>

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: f/centos76
Review: https://review.openstack.org/626688

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/626688
Committed: https://git.openstack.org/cgit/openstack/stx-integ/commit/?id=26cb76275997f060064d439e050384bead77b21b
Submitter: Zuul
Branch: f/centos76

commit acc1863b269fa974cd6c19b31c224dd88154e09d
Author: zhipengl <email address hidden>
Date: Tue Dec 11 00:30:56 2018 +0800

    Refactor source code patches for dhcp package

    3 source patches can be removed.
    2 patches adds support for wrs_install_uuid in the dhclient script.
    This added script part just copy the whole content of dhclient-enter-hooks.
    Following this script part, it will call this hook script if the hook
    exist under /etc/. However, our hook file existed in /etc/dhcp/ folder will
    be called by sbin/dhclient-script as well. I'd like to use dhcp config
    package to creat /etc/dhclient-enter-hooks soft linked to
    /etc/dhcp/dhclient-enter-hooks, so that it can call dhclient script and
    no need to add this 2 patches.

    Support-disable-nsupdate.patch can be removed as we already fixed port
    conflict issue in https://review.openstack.org/#/c/622711/

    Deployment test pass and related script file check pass!

    Story: 2004473
    Task: 28164

    Change-Id: If50ae697062a7d0c8a2831fbcc0f5641aaa41ec7
    Signed-off-by: zhipengl <email address hidden>

commit 61b8055a14f61851b9f70c76849bbb4f8f28ed55
Author: Steven Webster <email address hidden>
Date: Mon Dec 17 12:22:48 2018 -0500

    Fix remote logging traffic control filter priority

    Previous commit 01f5fdd made a required change to filter
    infrastructure traffic on the management interface with an 802.1q
    protocol in the case of a consolidated interface.

    However, this has caused the remote logging tc script to have a
    failure. The script tries to install 'ip' protocol filters at the
    same priority as the 802.1q filters, which is rejected by the
    kernel.

    This commit detects a consolidated interface situation and bumps
    the priority of the remote logging tc filter priority on the
    management interface, similarly to what is done in the main
    cgcs_tc_setup script.

    The file has also been cleaned up to pass bashate.

    Related-Bug: #1807055
    Change-Id: Id11625c0f9bcbf109f574563ff284d4a36bc6377
    Signed-off-by: Steven Webster <email address hidden>

commit 4dd1d96eddc84433ee3f6cf6f61db5b71a2d3b4c
Author: zhipengl <email address hidden>
Date: Sat Dec 15 01:34:18 2018 +0800

    Fix SFTP service is not working issue

    The root cause is that sftp path in sshd_config is not right.
    It should be changed from /usr/libexec/sftp-server
    to /usr/libexec/openssh/sftp-server

    Verified in my deployment environment
    sftp can connect to controller.

    Closes-Bug: 1808054

    Change-Id: Ia8d00abc1f18bc3b46faadd87f8ed153a446b7b0
    Signed-off-by: zhipengl <email address hidden>

commit 43514ea7fbd18d518511a165b59c82b7e20ebd8d
Author: Kwan, Louie <email address hidden>
Date: Wed Dec 12 15:54:30 2018 -0500

    [Enhancement] Add system active alarms in collect logs

    Currently the collect tool does not c...