Persistent/keepalive HTTP with the PHP Curl library?
PhpHttpCurlLibcurlKeep AlivePhp Problem Overview
I'm using a simple PHP library to add documents to a SOLR index, via HTTP.
There are 3 servers involved, currently:
- The PHP box running the indexing job
- A database box holding the data being indexed
- The solr box.
At 80 documents/sec (out of 1 million docs), I'm noticing an unusually high interrupt rate on the network interfaces on the PHP and solr boxes (2000/sec; what's more, the graphs are nearly identical -- when the interrupt rate on the PHP box spikes, it also spikes on the Solr box), but much less so on the database box (300/sec). I imagine this is simply because I open and reuse a single connection to the database server, but every single Solr request is currently opening a new HTTP connection via cURL, thanks to the way the Solr client library is written.
So, my question is:
- Can cURL be made to open a keepalive session?
- What does it take to reuse a connection? -- is it as simple as reusing the cURL handle resource?
- Do I need to set any special cURL options? (e.g. force HTTP 1.1?)
- Are there any gotchas with cURL keepalive connections? This script runs for hours at a time; will I be able to use a single connection, or will I need to periodically reconnect?
Php Solutions
Solution 1 - Php
cURL PHP documentation (curl_setopt) says:
> CURLOPT_FORBID_REUSE
- TRUE
to force
> the connection to explicitly close
> when it has finished processing, and
> not be pooled for reuse.
So:
- Yes, actually it should re-use connections by default, as long as you re-use the cURL handle.
- by default, cURL handles persistent connections by itself; should you need some special headers, check CURLOPT_HTTPHEADER
- the server may send a keep-alive timeout (with default Apache install, it is 15 seconds or 100 requests, whichever comes first) - but cURL will just open another connection when that happens.
Solution 2 - Php
Curl sends the keep-alive header by default, but:
- create a context using
curl_init()
without any parameters. - store the context in a scope where it will survive (not a local var)
- use
CURLOPT_URL
option to pass the url to the context - execute the request using
curl_exec()
- don't close the connection with
curl_close()
very basic example:
function get($url) {
global $context;
curl_setopt($context, CURLOPT_URL, $url);
return curl_exec($context);
}
$context = curl_init();
//multiple calls to get() here
curl_close($context);
Solution 3 - Php
-
On the server you are accessing keep-alive must be enabled and maximum keep-alive requests should be reasonable. In the case of Apache, refer to the apache docs.
-
You have to be re-using the same cURL context.
-
When configuring the cURL context, enable keep-alive with timeout in the header:
curl_setopt($curlHandle, CURLOPT_HTTPHEADER, array( 'Connection: Keep-Alive', 'Keep-Alive: 300' ));
Solution 4 - Php
If you don't care about the response from the request, you can do them asynchronously, but you run the risk of overloading your SOLR index. I doubt it though, SOLR is pretty damn quick.
https://stackoverflow.com/questions/124462/asynchronous-php-calls