Read tiles into an unzeroed buffer from the request pool, save ~0.6ms per tile

[leijurv]

Ever since #491, #492, tile_handler_serve's 1 megabyte buffer has been zero filled and pulled from a fresh pool with its own allocator, destroyed after every request. This means every tile request is going to fault in and zero 256 fresh pages of memory. This contends significantly under multithreaded load on the mmap lock. I have a test setup that mirrors tile.osm.org as well as I can (full planet, replaying tile requests, event MPM, 64 threads per child, HTTPS enabled, 28% of requests are If-None-Match conditional). Apache's CPU time per tile is decreased from 755 micros to 134 micro with this PR, which allocates the buffer from the request pool without zeroing it out. This matches the behavior of v0.8.1 which used plain malloc. The 134 micro time now matches v0.8.1 again.

[codecov]

Codecov Report

❌ Patch coverage is 50.00000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 82.05%. Comparing base (c4e3e62) to head (d0120af).
⚠️ Report is 45 commits behind head on master.

Files with missing lines	Patch %	Lines
src/mod_tile.c	50.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #517      +/-   ##
==========================================
+ Coverage   81.54%   82.05%   +0.51%     
==========================================
  Files          24       24              
  Lines        5645     5321     -324     
==========================================
- Hits         4603     4366     -237     
+ Misses       1042      955      -87     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[leijurv]

This one is actually not ready, as we can see from mac osx CI....

[leijurv]

Okay so actually the mac failure is due to a build time vs runtime discrepancy in apache portable runtime. mod_tile was built against APR 1.5.2, but the tests used homebrew's httpd 2.4.68 which linked against APR 1.7.6. This creates a strange situation where mod_tile's calls to apr_* are assuming 1.5.2 but the request pool objects were made by APR 1.7.6 from httpd. The allocator struct changed between those versions and a bunch of fields have different offsets which causes an immediate segfault on the first tile. Interestingly, smaller allocations are succeeding purely by chance because they only touch fields that didn't move.

[leijurv]

I was going to try to split this out into two or three stacked PRs but that would be confusing so I'll just update it here. We can fix the issue better by having APR's symbols bind to httpd's version at load time like how httpd's own symbols do already.