<html theme="default-dark" iconset="color"><head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
</head><body text="#000000"><br>
<br>
<span>Paul Ebersman wrote on 2021-10-05 13:23:</span><br>
<blockquote type="cite"
cite="mid:20211005202330.C6F8A15155FE@fafnir.remote.dragon.net">
<pre wrap="">fneves> ...
vcunat> I'm not sure if there can be *one* BCP way.
Definitely would need to be more a bag of tricks that operators can
mix/match based on their actual environment, customer base, etc.
Paid vs free probably have different concerns and obligations.
Folks with lots of smaller sites with lower qps rates per server vs
folks with a few much larger sites will have different pain points and
remediations.
I'd suspect that there are very few things that are always a good idea
for everyone everywhere.
...</pre>
</blockquote>
<br>
as a coauthor of resimprove, i suspect differently. some behaviours are
universally good, and these should be sought. what's at issue in the
cache miss data path is not what's good for a site or an operator, but
what's bad for the system as a whole. adding load to a nonresponsive
path is a universal badness, for example.<br>
<br>
if there are additional, more circumstantial, ways to handle cache miss
transactions, we should enumerate them. but, at heart, we have to
control our offered load, and we have the signal (timeouts) needed to do
so.<br>
<br>
<br>
<div class="moz-signature">-- <br>
<div>Sent from <a href="https://www.postbox-inc.com"><span style="color:
rgb(0, 157, 247);">Postbox</span></a></div></div>
</body></html>