Add docker-entrypoint-initdb.d support

[xzkostyan]

I hereby agree to the terms of the CLA available at: https://yandex.ru/legal/cla/?lang=en

This done in postgres entrypoint in the same way: https://github.com/docker-library/postgres/tree/master/11.

The most tricky part is starting clickhouse-server in detached mode and sleeping 1 second after it

[xzkostyan]

Solves #3319.

[xzkostyan]

As alternative port readiness check for the client can be performed in something like this:

while true; do
    timeout 1 bash -c 'cat < /dev/null > /dev/tcp/localhost/9000' 2>1 /dev/null;
    rv=$?
    if [[ $rv = 0 ]]; then
        break
    fi
done

I don't know which approach is better sleep or this cycle.

[proller]

Better use clickhouse-client

[xzkostyan]

There is no clickhouse-client binary inside the clickhouse-server image.

[filimonov]

And that is very confusing. IMHO client should be added to server container. It costs nothing (just a symlink) but makes usage of clickhouse-client much more logical.

[alexey-milovidov]

Don't understand, why we need array here.

[xzkostyan]

I see. In the postgres/mysql entrypoints password and database added later in script. This can be useful for future *_PASSWORD and *_USER variables support.

Array is redundant right now. I can change it to simple variable.

[alexey-milovidov]

I don't know which approach is better sleep or this cycle.

With current approach we have race condition.

while true; do
    timeout 1 bash -c 'cat < /dev/null > /dev/tcp/localhost/9000' 2>1 /dev/null;
    rv=$?
    if [[ $rv = 0 ]]; then
        break
    fi
done

With this approach we have possibility of infinite loop without any diagnostics.
Better to set some limit... for example, 10 retries.

[xzkostyan]

I can make another PR with port readiness probe within 10 retries. @alexey-milovidov should I?

[filimonov]

As alternative port readiness check for the client can be performed in something like this:

cat < /dev/null > /dev/tcp/localhost/9000

That will leave "warning" in the log file, something like that:

2018.12.10 12:08:49.548198 [ 25 ] {} <Trace> TCPHandlerFactory: TCP Request. Address: [::1]:58626
2018.12.10 12:08:49.549074 [ 25 ] {} <Warning> TCPHandler: Client has not sent any data.
2018.12.10 12:08:49.549131 [ 25 ] {} <Information> TCPHandler: Done processing connection.

Some users are quite nervous about such warnings... :\

In some scanrios (for example, if docker uses real network instead of NAT) it can give false match, or will never succeed. So port should not be hardcoded there.

Possible alternatives:

1. wait for a line "Ready for connections." in serverlog (something like that: https://superuser.com/a/449307 )
2. parametrize the port by something like ${CLICKHOUSE_PORT:-9000} (still will create a warning).
3. do the http ping wget --server-response --spider --quiet "http://localhost:8123" or echo 'HEAD /ping HTTP/1.0' > /dev/tcp/localhost/8123 (port still should be parametrized)

Also instead of parametrizing the port, you can extract it from config, like that:

clickhouse-extract-from-config --config-file /etc/clickhouse-server/config.xml  --key=tcp_port 
clickhouse-extract-from-config --config-file /etc/clickhouse-server/config.xml  --key=http_port 

summarize

For me that option looks the most straight ahead (easy to understand, no warnings, use wget possibilities to do retries with max number of retries, wget is avaliable in most of containers)

CLICKHOUSE_HTTP_PORT=$(clickhouse-extract-from-config --config-file /etc/clickhouse-server/config.xml  --key=http_port)
wget --server-response --spider --quiet --tries=20 --waitretry=1 --retry-connrefused "http://localhost:$CLICKHOUSE_HTTP_PORT/ping" && echo 'ok' || echo "ups"

[filimonov]

And also, while we are here - it's better to change those lines:
https://github.com/yandex/ClickHouse/blob/0466f6beafc7d1c0f20b4309e420ab3ab8b98a0f/docker/server/entrypoint.sh#L9-L15
to use clickhouse-extract-from-config