Description
After enabling the advanced HTTP cache via the --enable-http-cache flag, we observed an immediate and total drop in our 304 Not Modified HIT traffic.
When the flag is turned off, standard 304 behavior (both HIT and PASS) instantly returns. This behavior is fully reproducible across our deployment boundaries.
Expected Behavior
Enabling the HTTP cache should respect conditional request headers (such as If-None-Match and If-Modified-Since) and properly synthesize or pass through 304 Not Modified responses.
Actual Behavior
304 HIT traffic drops to near zero. Requests that should be served as a 304 are instead served as 200 OK HITs, or conditional revalidation fails entirely.
Production Evidence (Fastly Compute)
We monitored this behavior over a 5 day window.
- With flag ON: HIT 304 traffic dropped to near 0 across all routes.
- With flag OFF: HIT 304 traffic immediately returned to baseline levels.
Correlations: The traffic shifts align perfectly with our deployment boundaries.
Environment Details
Platform: Fastly Compute
Feature: --enable-http-cache flag
this was originally discussed here
Description
After enabling the advanced HTTP cache via the
--enable-http-cacheflag, we observed an immediate and total drop in our304 Not ModifiedHIT traffic.When the flag is turned off, standard 304 behavior (both HIT and PASS) instantly returns. This behavior is fully reproducible across our deployment boundaries.
Expected Behavior
Enabling the HTTP cache should respect conditional request headers (such as
If-None-MatchandIf-Modified-Since) and properly synthesize or pass through 304 Not Modified responses.Actual Behavior
304 HIT traffic drops to near zero. Requests that should be served as a 304 are instead served as 200 OK HITs, or conditional revalidation fails entirely.
Production Evidence (Fastly Compute)
We monitored this behavior over a 5 day window.
Correlations: The traffic shifts align perfectly with our deployment boundaries.
Environment Details
Platform: Fastly Compute
Feature: --enable-http-cache flag
this was originally discussed here