Fix parent pipeline number env on restarts

[qwerty287]

Closes #3680

[qwerty287]

What can we do about this failure? https://ci.woodpecker-ci.org/repos/3780/pipeline/15704/27#L183

[xoxys]

Not that simple. I had this issue quite often with Drone and Woodpecker while using the autoscaler. Summary from my observation:

• Some US cloud platforms do heavy IP based geo/- spam blocking
• With an autoscaler setup, you get a random IP from the Hetzner pool
• If a former owner of the IP has used it for bad habit, they get tainted and blocked by some providers

Solutions:

• Quick and dirty: Change the used Hetzer DC in the autoscaler deployment. The IP ranges seem to be different for the DCs, so temporarily switching to e.g. NUE, FSN or HEL worked for me in most cases.
• Use IPv6 only. It is quite difficult to perform IP-based filtering with IPv6. However, every component in the Woodpecker stack must support this. With Drone this was not possible, not sure about Woodpecker.
• Use an HTTP proxy server with a static IP and set HTTP_PROXY/HTTPS_PROXY env on the affected steps. However, most HTTP_PROXY servers cannot be used with authentication. Binding to a public IP allows anyone to use your proxy.
• Create a wiregurad network and connect all autoscaler hosts to this network. Use a persistent host with a static IP as a wiregurad gateway and route all Internet traffic via this gateway.

[anbraten]

Would caching using sth like the s3-cache help with this at least for the majority of executions? Otherwise as this problem seems to just occur rarely we are probably best to just throw away the agent and try again for now.

[xoxys]

Would caching using sth like the s3-cache help with this at least for the majority of executions

In many cases, no. It is in the nature of a CI system to contain dynamic parts and, at least in my experience, it is quite difficult to make pipelines work completely offline.

Otherwise as this problem seems to just occur rarely we are probably best to just throw away the agent and try again for now.

Keep in mind that the same IP address will be assigned to your machine again if it has not already been assigned to someone else. Just re-creating the agent might not work, that's why I had to switch the DC to ensure to get a new IP from another pool.

[qwerty287]

@anbraten Maybe you can just force-merge this for now?