Raise timeouts

Turns out latency is quite bad on the production server for
search.quarkus.io.

Author: yrodiere

_config.yml

@@ -28,16 +28,20 @@ search:
   min-chars: 2
   # The URL of the remote search service
   url: "https://search.quarkus.io/"
-  # The amount of time before we give up on a pending remote search and fall back to Javascript search
-  # The search service itself is rather fast (~10-30ms) but it's hosted in only one datacenter,
-  # so depending on how far from it you are, you can expect quite high latency...
-  # 500ms won't cut it for a DSL connection in France, for example.
-  initial-timeout: 1000
+  # The amount of time before we give up on a pending remote search and fall back to Javascript search.
+  # The search service itself is reasonably fast on a decent machine (with curl: ~100ms median, ~150ms 90th percentile).
+  # but it's slower on prod machines (with curl: ~200ms median, ~400ms 90th percentile),
+  # and prod network overhead makes it even worse (with curl from France: ~750ms median, ~1000ms 90th percentile),
+  # so we need a high timeout with some margin for varying network latency depending on browser location.
+  # Such a high timeout is acceptable since Javascript search is almost instantaneous,
+  # so when we hit the timeout the user won't experience much more delay.
+  # See also https://docs.google.com/spreadsheets/d/1w0tSfL-MKFArSrB-L8IX1pxmrhWmQyNgNwQRloG3cLk/edit#gid=456110917
+  initial-timeout: 1500
   # When fetching more pages, a timeout would be very bad for UX
   # since we would reset the whole page using Javascript --
   # while the user is reading through it!
   # This we set a higher timeout value, to make timeouts less likely.
-  more-timeout: 2000
+  more-timeout: 2500
 
 # Build settings
 error_mode: strict