🥇Nginx-log-collector utility from Avito for sending nginx logs to Clickhouse

This article will discuss the project nginx-log-collector, which will read nginx logs, send them to the Clickhouse cluster. Usually ElasticSearch is used for logs. Clickhouse requires fewer resources (disk space, RAM, CPU). Clickhouse writes data faster. Clickhouse compresses the data, which makes the data on disk even more compact. Benefits of Clickhouse can be seen in 2 slides from the report How VK inserts data into ClickHouse from tens of thousands of servers.

To view analytics by logs, let's create a dashboard for Grafana.

Who cares, welcome under cat.

Install nginx, grafana in the standard way.

Install clickhouse cluster with ansible-playbook from Denis Proskurin.

Creating a database and tables in Clickhouse

In this file SQL queries for creating databases and tables for nginx-log-collector in Clickhouse are described.

We make each request in turn on each server of the Clickhouse cluster.

Important note. In this line, logs_cluster should be replaced with your cluster name from the clickhouse_remote_servers.xml file between "remote_servers" and "shard".

ENGINE = Distributed('logs_cluster', 'nginx', 'access_log_shard', rand())

Installing and configuring nginx-log-collector-rpm

Nginx-log-collector does not have an rpm. Here https://github.com/patsevanton/nginx-log-collector-rpm create rpm for it. rpm will be built using Fedora Copr

Install rpm package nginx-log-collector-rpm

yum -y install yum-plugin-copr
yum copr enable antonpatsev/nginx-log-collector-rpm
yum -y install nginx-log-collector
systemctl start nginx-log-collector

Edit config /etc/nginx-log-collector/config.yaml:

  .......
  upload:
    table: nginx.access_log
    dsn: http://ip-адрес-кластера-clickhouse:8123/

- tag: "nginx_error:"
  format: error  # access | error
  buffer_size: 1048576
  upload:
    table: nginx.error_log
    dsn: http://ip-адрес-кластера-clickhouse:8123/

Setting up nginx

General nginx config:

user  nginx;
worker_processes  auto;

#error_log  /var/log/nginx/error.log warn;
pid        /var/run/nginx.pid;

events {
    worker_connections  1024;
}

http {
    include       /etc/nginx/mime.types;
    default_type  application/octet-stream;

    log_format  main  '$remote_addr - $remote_user [$time_local] "$request" '
                      '$status $body_bytes_sent "$http_referer" '
                      '"$http_user_agent" "$http_x_forwarded_for"';

    log_format avito_json escape=json
                     '{'
                     '"event_datetime": "$time_iso8601", '
                     '"server_name": "$server_name", '
                     '"remote_addr": "$remote_addr", '
                     '"remote_user": "$remote_user", '
                     '"http_x_real_ip": "$http_x_real_ip", '
                     '"status": "$status", '
                     '"scheme": "$scheme", '
                     '"request_method": "$request_method", '
                     '"request_uri": "$request_uri", '
                     '"server_protocol": "$server_protocol", '
                     '"body_bytes_sent": $body_bytes_sent, '
                     '"http_referer": "$http_referer", '
                     '"http_user_agent": "$http_user_agent", '
                     '"request_bytes": "$request_length", '
                     '"request_time": "$request_time", '
                     '"upstream_addr": "$upstream_addr", '
                     '"upstream_response_time": "$upstream_response_time", '
                     '"hostname": "$hostname", '
                     '"host": "$host"'
                     '}';

    access_log     syslog_server=unix:/var/run/nginx_log.sock,nohostname,tag=nginx avito_json; #ClickHouse
    error_log      syslog_server=unix:/var/run/nginx_log.sock,nohostname,tag=nginx_error; #ClickHouse

    #access_log  /var/log/nginx/access.log  main;

    proxy_ignore_client_abort on;
    sendfile        on;
    keepalive_timeout  65;
    include /etc/nginx/conf.d/*.conf;
}

Virtual host one:

vhost1.conf:

upstream backend {
    server ip-адрес-сервера-с-stub_http_server:8080;
    server ip-адрес-сервера-с-stub_http_server:8080;
    server ip-адрес-сервера-с-stub_http_server:8080;
    server ip-адрес-сервера-с-stub_http_server:8080;
    server ip-адрес-сервера-с-stub_http_server:8080;
}

server {
    listen   80;
    server_name vhost1;
    location / {
        proxy_pass http://backend;
    }
}

Add virtual hosts to the /etc/hosts file:

ip-адрес-сервера-с-nginx vhost1

HTTP server emulator

As an HTTP server emulator we will use nodejs-stub-server from Maxim Ignatenko

nodejs-stub-server doesn't have rpm. Here https://github.com/patsevanton/nodejs-stub-server create rpm for it. rpm will be built using Fedora Copr

Install nodejs-stub-server package on upstream nginx rpm

yum -y install yum-plugin-copr
yum copr enable antonpatsev/nodejs-stub-server
yum -y install stub_http_server
systemctl start stub_http_server

Stress Testing

Testing is carried out using Apache benchmark.

Install it:

yum install -y httpd-tools

We start testing using Apache benchmark from 5 different servers:

while true; do ab -H "User-Agent: 1server" -c 10 -n 10 -t 10 http://vhost1/; sleep 1; done
while true; do ab -H "User-Agent: 2server" -c 10 -n 10 -t 10 http://vhost1/; sleep 1; done
while true; do ab -H "User-Agent: 3server" -c 10 -n 10 -t 10 http://vhost1/; sleep 1; done
while true; do ab -H "User-Agent: 4server" -c 10 -n 10 -t 10 http://vhost1/; sleep 1; done
while true; do ab -H "User-Agent: 5server" -c 10 -n 10 -t 10 http://vhost1/; sleep 1; done

Setting up Grafana

You will not find a dashboard on the official Grafana website.

Therefore, we will do it by hand.

You can find my saved dashboard here.

You also need to create a table variable with the content nginx.access_log.

Singlestat Total Requests:

SELECT
 1 as t,
 count(*) as c
 FROM $table
 WHERE $timeFilter GROUP BY t

Singlestat Failed Requests:

SELECT
 1 as t,
 count(*) as c
 FROM $table
 WHERE $timeFilter AND status NOT IN (200, 201, 401) GROUP BY t

Singlestat Failing Percent:

SELECT
 1 as t, (sum(status = 500 or status = 499)/sum(status = 200 or status = 201 or status = 401))*100 FROM $table
 WHERE $timeFilter GROUP BY t

Singlestat Avg Response Time:

SELECT
 1, avg(request_time) FROM $table
 WHERE $timeFilter GROUP BY 1

Singlestat Max Response Time:

SELECT
 1 as t, max(request_time) as c
 FROM $table
 WHERE $timeFilter GROUP BY t

Count Status:

$columns(status, count(*) as c) from $table

To output data like a pie, you need to install the plugin and reload grafana.

grafana-cli plugins install grafana-piechart-panel
service grafana-server restart

Pie TOP 5 Status:

SELECT
    1, /* fake timestamp value */
    status,
    sum(status) AS Reqs
FROM $table
WHERE $timeFilter
GROUP BY status
ORDER BY Reqs desc
LIMIT 5

Further I will give requests without screenshots:

Count http_user_agent:

$columns(http_user_agent, count(*) c) FROM $table

GoodRate/BadRate:

$rate(countIf(status = 200) AS good, countIf(status != 200) AS bad) FROM $table

response timing:

$rate(avg(request_time) as request_time) FROM $table

Upstream response time (response time of the 1st upstream):

$rate(avg(arrayElement(upstream_response_time,1)) as upstream_response_time) FROM $table

Table Count Status for all vhosts:

$columns(status, count(*) as c) from $table

General view of the dashboard

Comparing avg() and quantile()

avg()

quantile()

Conclusion:

Hopefully the community will get involved in developing/testing and using nginx-log-collector.
And when someone implements nginx-log-collector, he will tell you how much he saved disk, RAM, CPU.

Telegram channels:

Milliseconds:

Who cares about milliseconds, write or vote, please, in this issue.

Source: habr.com

Nginx-log-collector utility from Avito for sending nginx logs to Clickhouse