It turns out no browsers implement HTTP/2 GOAWAY handling properly, and large enough number of resources on a page results in failures to load some resources. In particular, Chrome seems to experience errors if loading of all resources requires more than 1 connection (while it is usually able to retry requests at least once, even with 2 connections there are occasional failures for some reason), Safari if loading requires more than 3 connections, and Firefox if loading requires more than 10 connections (can be configured with network.http.request.max-attempts, defaults to 10). It does not seem to be possible to resolve this on nginx side, even strict limiting of maximum concurrency does not help, and loading issues seems to be triggered by merely queueing of a request for a particular connection. The only available mitigation seems to use higher keepalive_requests value. The new default is 1000 and matches previously used default for http2_max_requests. It is expected to be enough for 99.98% of the pages (https://httparchive.org/reports/state-of-the-web?start=latest#reqTotal) even in Chrome.
Similar to lingering_time, it limits total connection lifetime before keepalive is switched off. The default is 1 hour, which is close to the total maximum connection lifetime possible with default keepalive_requests and keepalive_timeout.
In current versions (all versions based on zlib 1.2.11, at least since 2018) it no longer uses 64K hash and does not force window bits to 13 if it is less than 13. That is, it needs just 16 bytes more memory than normal zlib, so these bytes are simply added to the normal size calculation.
In limit_req, auth_delay, and upstream code to check for broken connections, tests for possible connection close by the client did not work if the connection was already closed when relevant event handler was set. This happened because there were no additional events in case of edge-triggered event methods, and read events were disabled in case of level-triggered ones. Fix is to explicitly post a read event if the c->read->ready flag is set.
The response size check introduced in 39501ce97e29 did not take into account possible padding on DATA frames, resulting in incorrect "upstream sent response body larger than indicated content length" errors if upstream server used padding in responses with known length. Fix is to check the actual size of response buffers produced by the code, similarly to how it is done in other protocols, instead of checking the size of DATA frames. Reported at: http://mailman.nginx.org/pipermail/nginx-devel/2021-March/013907.html
Sun C complains about "statement not reached" if a "return" is followed by additional statements.
The stub status module and ngx_http_send_response() (used by the empty gif
module and the "return" directive) incorrectly assumed that responding
to HEAD requests always results in r->header_only being set. This is not
true, and results in incorrect behaviour, for example, in the following
configuration:
location / {
image_filter size;
return 200 test;
}
Fix is to remove this incorrect micro-optimization from both stub status
module and ngx_http_send_response().
Reported by Chris Newton.
After 7675:9afa45068b8f and 7678:bffcc5af1d72 (1.19.1), during non-buffered simple proxying, responses with extra data might result in zero size buffers being generated and "zero size buf" alerts in writer. This bug is similar to the one with FastCGI proxying fixed in 7689:da8d758aabeb. In non-buffered mode, normally the filter function is not called if u->length is already 0, since u->length is checked after each call of the filter function. There is a case when this can happen though: if the response length is 0, and there are pre-read response body data left after reading response headers. As such, a check for u->length is needed at the start of non-buffered filter functions, similar to the one for p->length present in buffered filter functions. Appropriate checks added to the existing non-buffered copy filters in the upstream (used by scgi and uwsgi proxying) and proxy modules.