MariaDB/MySQL slow query log to syslog (and other output modules)

MariaDB (fork of MySQL), with a few lines of configuration, provide a very interesting log regarding application-level query performances: the slow-logs.
Sadly, and for a reason I have yet to discover, these logs are not syslog-enabled. Thus, you cannot beneficiate of all the advantages of using the syslog protocol, ie: centralization, backup/archive, parsing, alerting… To make things even easier, these very interesting logs are multi-lines…
In a company where application developers are interested by the performances of their application and to optimize their code and queries, receiving information about the production application-level slow queries is necessary.
Here’s a sample mariadb/rsyslog configuration that achieve slow-logs to syslog and also generates a snmptrap per slow query (you could use the smtp output module instead or any other rsyslog output module for what matters).

  • /etc/my.cnf.d/logging.cnf

# Slow query Logging Settings
  • /etc/rsyslog.d/zz-mariadb.conf
# MariaDB-Syslog Config File

if $app-name == 'mysqld' then /var/log/mariadb/mariadb.log
& stop

if $app-name == 'mysqld-scl-helper' then /var/log/mariadb/mariadb.log
& stop

# Slow logs (not syslog-enabled)
module(load="imfile" PollingInterval="10") # once
    # Because logs are multi-line
    # Time: 180618 16:23:33
    # Time: 180618 8:43:16
    startmsg.regex="^# Time: [[:digit:]]{6} "
    # Remove LF escaping so that it's properly formatted within Castlerock
  • /etc/rsyslog.d/zz-mariadb-slowq-to-trap.conf
# Send SNMP traps for slow logs
module(load="omsnmp") # once
$actionsnmptransport udp
$actionsnmptarget <snmp_server_ip_addr_or_fqdn>
$actionsnmptargetport 162
$actionsnmpversion 1 # means SNMPv2c
$actionsnmpcommunity <snmp_community>
$actionsnmptrapoid <snmp_oid>

if $app-name == 'mysqld-slowlog' then :omsnmp:
& stop

Proper end-to-end syslog logging with Apache httpd

Table of content

  • Context
  • Objectives
  • Folder structures and relevant files
  • Configurations
    • /etc/httpd/conf.d/httpd_base_module.conf
    • /etc/httpd/conf.d/
    • /opt/fcrouzat/syslog/
    • /etc/rsyslog.d/zz-10-httpd.conf
    • Remote logging: /etc/rsyslog.d/zz-09-remote-httpd.conf
  • Caveats
  • Conclusions


For an unknown reason, Apache httpd only supports syslog since 2.4 and only for error logs.
On top of that, default log format is very poor and (to me) doesn’t mean anything.
Add these two informations and you’ll give any sysadmin a headache.

Fortunately there is a way around these issues by clever usage of the named pipes. This W/A is not so complicated but requires to be careful and a properly configured stack.


Using this workaround I will demonstrate how to perform proper end-to-end syslog logging with both httpd 2.2 and 2.4.
This is very helpful in PCI-DSS compliant environment because it allow you to easily centralize Apache httpd logs. As a reminder: unlike many other regulations and standards, PCI DSS explicitly requires organization to implement log management. Requirement 10 explicitly requires organizations stating: logging mechanisms and the ability to track user activities are critical in preventing, detecting, or minimizing the impact of a data compromise. The presence of logs in all environments allows thorough tracking, alerting, and analysis when something does go wrong. Determining the cause of a compromise is very difficult, if not impossible, without system activity logs.[1]

In addition to proper logging through syslog (both local and to remote hosts), and since we are about to modify our Apache httpd configuration for the better, we will use this opportunity to improve various things that always bugged me with httpd: having a maintained and useful LogFormat and having proper filenames for the local log files.

Testing and validation have been done under the following configurations:

  • RHEL 6.x + httpd-2.2
  • RHEL 7.x + httpd-2.4

Depending on how you might have already tweaked your LogFormat, your logs should looks like this: - - [21/Jan/2016:16:42:25 +0100] "GET /wp-content/ HTTP/1.1" 200 13776 "" "Mozilla/5.0 (Windows NT 6.1; WOW64; rv:43.0) Gecko/20100101 Firefox/43.0"

It is not syslog-compliant, date format is quite unexpected, there is no app-name, and it is probably missing interesting informations. In addition, this is for example the combined LogFormat, but there are others and they are all different, you cannot find a discriminating way to filter them or manipulate them: remember, no app-name …
Assuming you’d like to parse or analyze these logs, you’d have to write very specific decoder because you cannot just use a filter on the app-name. The same goes for sending these logs to remote hosts over syslog, you cannot easily filter to only send Apache httpd logs without writing complex decoders…

We want at least to do the following: remove the poorly written datetime and add a valid syslog header with a deterministic app-name in front of the untouched message. A log line would then look like this:

Jan 15 03:28:07 httpd: - - "GET /wp-content/ HTTP/1.1" 200 13776 "" "Mozilla/5.0 (Windows NT 6.1; WOW64; rv:43.0) Gecko/20100101 Firefox/43.0"

But while doing so, why not also improve everything by replacing the default LogFormat (combined) with one of our own that is a lot more powerful. Of course, feel free to adapt my “powerful” LogFormat to your needs. A log line would then look like this:

Jan 15 03:28:07 httpd: TLSv1.2(ECDHE-RSA-AES256-GCM-SHA384) "GET /index.html HTTP/1.1" 200 11301B "- - -"

Folder structures and relevant files

You can obviously organize your folder structure as you wish and name your files as you intend to. But, to be consistent I’ll provide the filenames I have been using.
All my configuration is done in /etc/httpd/conf.d/ and all *.conf files are included from httpd’ default configuration /etc/httpd/conf/httpd.conf. I never edit this package-shipped file. Same applies for rsyslog with /etc/rsyslog.d/ included from package-shipped /etc/rsyslog.conf

Here are the relevant files we are going to use:

  • /etc/httpd/conf.d/httpd_base_module.conf : define LogFormat and stuff module dependent
  • /etc/httpd/conf.d/ : virtualhost file for
  • /etc/rsyslog.d/zz-10-httpd.conf : define rsyslog httpd-related configurations
  • /opt/fcrouzat/syslog/ : syslog wrapper for httpd



First, create a new LogFormat that match your needs, feel free to define your own but do not includes date, time or app-name. This has to be the raw message. The syslog header will be added later.

    LogFormat "%h:%{remote}p %{Host}i:%{canonical}p %{SSL_PROTOCOL}x(%{SSL_CIPHER}x) \"%m %U %H\" %>s %BB \"%!200,302,304{Referer}i %!200,302,304{User-Agent}i %!200,302,304{cookie}i\"" default


We have to modify our virtualhost(s) to use the new LogFormat and also the named pipe.
Apache httpd must never write directly to disk anymore, everything has to be piped.
This named pipe execute a custom-made script that takes 3 arguments:

  • Virtualhost FQDN
  • Log type: value can be access or error
  • Encryption: value can be http or https
    CustomLog "|/opt/fcrouzat/syslog/ access https" default
    ErrorLog "|/opt/fcrouzat/syslog/ error https"


This is the wrapper used as a named pipe by httpd. Every single line of log is piped to this script via stdout.
This script uses logger to address rsyslog.
Argument 2 is used to determine if we are handling access or error logs.
Argument 3 is used to determine if we are handling http or https logs.
Argument 1 is passed after the fixed part to have the knowledge of the virtualhost name in rsyslog as you’ll see after
This is useful to use because we will use that to define filename pattern for our logs such as:

  • /var/log/httpd/<fqdn>.access.log
  • /var/log/httpd/<fqdn>.error.log
  • /var/log/httpd/<fqdn>.ssl.access.log
  • /var/log/httpd/<fqdn>.ssl.error.log

This is purely cosmetic and is not really syslog-related but having properly named logfile has shown to be useful.


# Log through syslog using logger and set the "app-name" protocol header with
# "httpd-(access|error)-http[s]-".

while read line ; do
    [[ $2 == access ]] && [[ $3 == http ]]  && /bin/logger -p  -t "httpd-access-http-${1}"
    [[ $2 == access ]] && [[ $3 == https ]] && /bin/logger -p  -t "httpd-access-https-${1}"
    [[ $2 == error ]]  && [[ $3 == http ]]  && /bin/logger -p daemon.error -t "httpd-error-http-${1}"
    [[ $2 == error ]]  && [[ $3 == https ]] && /bin/logger -p daemon.error -t "httpd-error-https-${1}"


Here’s the deal: we now have httpd logging through a named piped that uses logger to perform syslog. We also have relevant informations from the app-name that has been set with the -t flag from logger. We can use all these properties to add the relevant syslog header on httpd raw message and log to a properly named logfile.

$PreserveFQDN on

# Log Format template: add a proper syslog header and append %msg%

$template HttpdFormat,"%timegenerated% %hostname% httpd: %msg%\n"
$template HttpdRemoteFormat,"<%PRI%>%timegenerated% %hostname% httpd: %msg%\n"

# Dynamic app-name logfile templates

# Remove "httpd-access-http-" from %app-name% to extract the virtualhost name
$template HttpAccessLog,"/var/log/httpd/%app-name:19:$:%.access.log"
# Remove "httpd-access-https-" from %app-name% to extract the virtualhost name
$template HttpsAccessLog,"/var/log/httpd/%app-name:20:$:%.ssl.access.log"
# Remove "httpd-error-http-" from %app-name% to extract the virtualhost name
$template HttpErrorLog,"/var/log/httpd/%app-name:18:$:%.error.log"
# Remove "httpd-error-https-" from %app-name% to extract the virtualhost name
$template HttpsErrorLog,"/var/log/httpd/%app-name:19:$:%.ssl.error.log"

# Syslog appname-based routing with proper template depending on access/error and http/https

if $app-name startswith 'httpd-access-http-' then ?HttpAccessLog;HttpdFormat
& stop

if $app-name startswith 'httpd-access-https-' then ?HttpsAccessLog;HttpdFormat
& stop

if $app-name startswith 'httpd-error-http-' then ?HttpErrorLog;HttpdFormat
& stop

if $app-name startswith 'httpd-error-https-' then ?HttpsErrorLog;HttpdFormat
& stop

Remote logging: /etc/rsyslog.d/zz-09-remote-httpd.conf

If you want to forward httpd logs to a remote host now that they are properly formatted, just add the following configuration prior to the & stop instructions, hence the numbering on the filename. We are also using a dedicated syslog template for remote messages that preserve the syslog priority by using <%PRI%>.
This setup is using TCP syslog forward with failover.

if $app-name startswith 'httpd-' then {
    $ActionExecOnlyWhenPreviousIsSuspended on
    $ActionExecOnlyWhenPreviousIsSuspended off
# DEFAULT (all)
else {
    $ActionExecOnlyWhenPreviousIsSuspended on
    $ActionExecOnlyWhenPreviousIsSuspended off


There is at least one caveat with this setup: for an unknown reason, with RHEL6+httpd-2.2 there is an extra leading space in the %msg% part of the log when piped. Because of that, you have to replace httpd: %msg%\n" with httpd:%msg%\n" in the LogFormat section of the rsyslog configuration to have a properly formatted message on the wire and in the logs.


Besides TL;DR and for those who survived, as we have seen in the introduction, having a proper end-to-end configuration for httpd and syslog is not available out-of-the-box and requires many adjustments and maybe skills. Thanks to this setup we have been able to make httpd log through syslog and also do a little cleanup of the configuration: we have properly named logfiles and relevant content, we can filter locally or remotely on the app-name and write decoders from there.

As for stressing this setup: I am currently running this on different servers with approx. 2,000,000 lines of log every day each and without any latency on RHEL7.x + httpd-2.4 and SELinux=enforcing.

[1] –

Cisco AAA using TACACS+ with syslog and PAM support for CentOS/RedHat

Table of content

  • Context
  • TACACS+ Server installation
  • TACACS+ Server Configuration
  • Running TACACS+


In order to provide centralized authentication to network devices, radius is commonly used and works very well. It is generic and be used as a “proxy” to any kind of authentication backend which is great. If you have a very strict security policy or are running network devices in a very strict environment (PCI-DSS is a good example) it is mandatory to provides audit logs of all actions executed by privileged users and if possible, by anyone.

If ever your company is working with Cisco equipments, then it appears it is not possible to use radius for accounting and you have to use … TACACS… While you are stuck with a proprietary solution, the “good” news is that you can plug TACACS with PAM and use it for every A in AAA: authentication, authorization and accouting. Here’s how:

TACACS+ Server installation

It is important to have pam-devel installed during the compilation so that tac_plus is compiled with PAM support.
Verify in your ./configure output that PAM support has been enabled.

cd /usr/local/src/
tar xvzf tacacs+-F4.0.4.27a.tar.gz
cd tacacs+-F4.0.4.27a

yum install flex bison tcp_wrappers-devel tcp_wrappers-lib gcc pam-devel

./configure --prefix=/opt/tacacs+
make install
chmod 750 /opt/tacacs+
mkdir /opt/tacacs+/etc/

yum remove flex bison tcp_wrappers-devel tcp_wrappers-lib gcc cpp mpfr ppl cloog-ppl pam-devel

TACACS+ Server Configuration

  • /opt/tacacs+/etc/tac_plus.conf:

accounting syslog
logging = local2

# read only group
# not sure it works, haven't used it yet
group = readonly {
    default service = deny
    service = exec {
            priv-lvl = 0
    cmd=show {
            permit .*
    cmd=enable {
                permit .*
    cmd=exit {
                permit .*

# admin group
group = admins {
        default service = permit
        login = PAM
        service = exec {
             priv-lvl = 15

# Create a block for every admin user you have
user = fcrouzat {
        member = admins
  • /etc/pam.d/tac_plus:
    auth       include      system-auth
    account    required
    account    include      system-auth
    password   include      system-auth
    session    optional force revoke
    session    include      system-auth
    session    required

Next step is to open firewalling between your Cisco(s) and your TACACS+ server(s) for the TCP/49 port.
Now back to the system, we must configuration rsyslogd /etc/rsyslog.d/tacacs.conf

if $app-name == 'tac_plus' then /var/log/tacacs.log
& stop

… and logrotate /etc/logrotate.d/tacacs

/var/log/tacacs.log {
    rotate 4
    create 0640 root root
        /bin/kill -HUP `cat /var/run/ 2>/dev/null` 2>/dev/null || true

Client configuration (Cisco)

! use tacacs+ for authentication
aaa authentication login default local group tacacs+
! authorization if authenticated hence with tacacs+
aaa authorization exec default if-authenticated

! accounting with tacacs+
aaa accounting commands 0 default
 action-type start-stop
 group tacacs+
! accounting with tacacs+
aaa accounting commands 1 default
 action-type start-stop
 group tacacs+
! accounting with tacacs+
aaa accounting commands 15 default
 action-type start-stop
 group tacacs+
! tacacs+ server configuration
tacacs server logrelay01xxx
 address ipv4 192.168.XX.YY

Running TACACS+

Finally you can run the server in debug mode:

/opt/tacacs+/bin/tac_plus -C /opt/tacacs+/etc/tac_plus.conf -L -p 49 -d128 -g

… or use the very basic following initscript[1] which works for EL6:

wget -O /etc/init.d/tac_plus
restorecon -RvF /etc/init.d/tac_plus
chmod 755 /etc/init.d/tac_plus
chkconfig tac_plus on
chkconfig --list tac_plus
service tac_plus start

Finally, you can tail the TACACS+ server logs and complete the setup by logging on a Cisco and verify that authentication happen, then authorization and finally accounting are working:

Feb 22 15:38:13 tac_plus[25833]:    fcrouzat    tty1    stop    task_id=124    timezone=UTC    service=shell    start_time=1456152071    priv-lvl=1    cmd=show vlan brief 
Feb 22 15:38:17 tac_plus[25834]: connect from []
Feb 22 15:38:17 tac_plus[25834]:    fcrouzat    tty1    stop    task_id=125    timezone=UTC    service=shell    start_time=1456152076    priv-lvl=1    cmd=show interfaces trunk 

[1] –