We've tended to disable backend servers for specific backends as part of our deployer integration, so it makes a certain amount of sense for us to use the same basic methodology to push something into a particular backend. (We also don't have to restart haproxy to adjust the configuration, since the admin/stats socket allows dynamic "MAINT" mode.)
In case it's useful for anyone, the CLI setup for dealing with this in batch (nbproc > 1 with an N+1 cluster of haproxy servers) is here: https://github.com/jbuchbinder/hactl I have similar logic built into our software deployer to handle it transparently. (I'm not sure if the "backup" attribute for backend servers is *supposed* to be used the way we've been using it, but it does a great job of routing traffic elsewhere when all servers are either down or in MAINT.) Thanks, Jeff ________________________________________ From: Philipp Buehler <[email protected]> Sent: Sunday, January 24, 2016 12:20 PM To: Jeff Buchbinder Cc: Haproxy Subject: Re: Temporary Maintenance frontend for all port 80/443 ? Am 24.01.2016 17:58 schrieb Jeff Buchbinder: > The way we've done it has been to add a maintenance server to all of > our backends, then use the admin socket to iteratively disable all > "real" servers for our backends. The maintenance server points all > requests to your temporary page. > > Something like: > > server maintenance 10.0.5.10:80 backup check inter 15000 rise 2 fall > 5 We're doing yet another approach, which is easier to "steer". Watch closely about the logic .. frontend whatever acl usrnb nbsrv(usr) gt 0 acl usr_h_nb nbsrv(usr_h) eq 0 use_backend usr if usrnb usr_h_nb use_backend usr_h backend usr server usr usr021:8080 check downinter 30000 fastinter 3000 fall 2 backend usr_h server helper localhost:8000 check downinter 30000 fastinter 3000 fall 2 option httpchk GET /maint.txt HTTP/1.1\r\nHost:\ hidden - NO need to fiddle with haproxy.conf and/or restarts of haproxy itself - regular ops: usrnb is true, usr_h_nb is false (no maint.txt/nginx) - irr ops: usrnb is false -> use usr_h (which has its own logic for serving sorry pages) - maint: "touch maint.txt" -> usr_h_nb is false -> use usr_h (value of usrnb irrelevant) FTR; nginx.conf snippet: server { listen 8000; server_name hidden; access_log /var/log/nginx/maint.log main; location /maint.txt { access_log off; } location / { sorry-pages-things; } } HTH, -- pb
