docs/faq.md
No. Guzzle can use any HTTP handler to send requests. This means that Guzzle can be used with cURL, PHP's stream wrapper, sockets, and non-blocking libraries like React. You just need to configure an HTTP handler to use a different method of sending requests.
[!NOTE] Guzzle has historically only utilized cURL to send HTTP requests. cURL is an amazing HTTP client (arguably the best), and Guzzle will continue to use it by default when it is available. It is rare, but some developers don't have cURL installed on their systems or run into version specific issues. By allowing swappable HTTP handlers, Guzzle is now much more customizable and able to adapt to fit the needs of more developers.
Yes. You can use the requestAsync, sendAsync, getAsync, headAsync,
putAsync, postAsync, deleteAsync, and patchAsync methods of a client to
send an asynchronous request. For asynchronous requests that do not have a named
shortcut method, such as OPTIONS requests, use requestAsync() with the method
name. The client will return a
GuzzleHttp\Promise\PromiseInterface<Psr\Http\Message\ResponseInterface, mixed>
object. You can chain then functions off of the promise for fulfilled
responses and rejected reasons.
$promise = $client->requestAsync('GET', 'http://httpbin.org/get');
$promise->then(function ($response) {
echo 'Got a response! ' . $response->getStatusCode();
}, function ($reason) {
// The rejection reason is often a Guzzle exception, but custom handlers can
// reject with other values.
});
You can force an asynchronous response to complete using the wait() method of
the returned promise. It returns the response on fulfillment and throws when the
promise is rejected.
$promise = $client->requestAsync('GET', 'http://httpbin.org/get');
$response = $promise->wait();
cURL offers a huge number of
customizable options. While Guzzle
normalizes many of these options across different handlers, there are times
when you need to set custom cURL options. This can be accomplished by passing
an array keyed by allow-listed integer CURLOPT_* constants in the curl
key of a request. Raw cURL options outside the built-in cURL handlers'
allow-list are rejected. The special body_as_string key is also recognized
by Guzzle's cURL handler.
For example, let's say you need to customize the outgoing network interface used with a client.
$client->request('GET', '/', [
'curl' => [
CURLOPT_INTERFACE => 'xxx.xxx.xxx.xxx'
]
]);
Custom cURL request options remain active during redirects unless Guzzle
documents otherwise. See allow_redirects
for cross-origin redirect credential behavior.
Callbacks supplied directly through the curl request option are passed to
PHP's cURL extension as low-level callbacks. Guzzle does not normalize exception
or abort behavior for raw cURL callbacks. Prefer Guzzle's progress,
on_headers, and on_stats request options when you want Guzzle's documented
callback semantics.
If you use asynchronous requests with the cURL multi handler, the client can bound concurrent connections with named constructor options:
use GuzzleHttp\Client;
$client = new Client([
'max_total_connections' => 50,
'max_host_connections' => 5,
]);
When constructing a cURL multi handler yourself, pass the same named options to the handler:
use GuzzleHttp\Client;
use GuzzleHttp\Handler\CurlMultiHandler;
use GuzzleHttp\HandlerStack;
$handler = new CurlMultiHandler([
'max_total_connections' => 50,
'max_host_connections' => 5,
]);
$client = new Client(['handler' => HandlerStack::create($handler)]);
Additional cURL multi options that do not have named Guzzle options can still be
specified as an array keyed by integer CURLMOPT_* constants in the options
key of the CurlMultiHandler constructor. For example,
CURLMOPT_MAX_CONCURRENT_STREAMS can be used on PHP versions that expose it.
Multiplexing on the multi handle is controlled by the named multiplex option
rather than a raw cURL multi option: pass Multiplexing::NONE as the
multiplex client option or, when constructing the handler yourself, to the
CurlMultiHandler constructor to disallow multiplexing for every transfer the
handler runs. Multiplexing::NONE is also accepted as a request option value
exactly where its guarantee - the transfer does not share its connection with
any concurrent transfer - holds and can be verified. Passing
CURLMOPT_PIPELINING in the options array is rejected.
Numeric connection caps are enforced by CurlMultiHandler. When the caps are
configured, the default handler routes synchronous requests through the capped
CurlMultiHandler as well, and a cap-configured fallback StreamHandler
rejects enabled response streaming (stream => true) because streamed
connections cannot be capped. Accepted stream-handler transfers are buffered and
hold at most one connection per in-flight call, including in stream fallback
environments without a cap-capable cURL where every request uses the stream
handler. Overlapping buffered calls are not collectively limited by the
configured numbers.
Manually constructed CurlHandler or custom handlers are outside these caps. A
custom handle_factory is likewise caller-controlled and must not attach an
external connection-sharing CURLOPT_SHARE pool when the caps must hold,
because Guzzle cannot inspect that native handle state.
Connection cap options compose with transport sharing as follows. Handler
transport sharing shares only DNS and, when supported, TLS session data and
works with the caps unchanged. Persistent transport sharing also pools
connections in a shared cURL share handle, and libcurl does not apply the cURL
multi connection cap options to transfers that use a shared connection pool.
When the caps are configured, TransportSharing::PERSISTENT_PREFER therefore
falls back to handler-lifetime sharing, and
TransportSharing::PERSISTENT_REQUIRE is rejected.
The caps bound open connections, including idle pooled connections, rather than in-flight requests.
max_host_connections follows libcurl's connection-bundle grouping. Proxy
forwarding, CONNECT tunnels, SOCKS proxies, and older libcurl versions do not
necessarily group connections under the same host key, so it is not a portable
per-proxy or per-credential socket limit.
At the limit, libcurl queues transfers waiting for a connection slot. On libcurl
8.8.0 and newer, transfers queued behind a cap keep consuming the request
timeout; older libcurl does not run timeout checks on queued transfers, so a
queued transfer can overstay its timeout until a connection slot frees (curl
issue #13276). On libcurl 8.16.0 through 8.19.x, queued transfers resume one at
a time in an order that can starve some transfers under sustained load (fixed in
8.20.0). A transfer that must open a fresh connection, such as a proxied
transfer Guzzle isolates for credential safety, does not bypass the caps: it can
evict an eligible idle connection, and otherwise it stays pending until the
applicable per-host or total cap frees.
Low caps combined with aggressive timeouts and large request bursts can
therefore time out, or wait longer than expected, before a connection becomes
available. To bound in-flight requests and memory, combine the caps with
request-level concurrency controls such as GuzzleHttp\Pool or
GuzzleHttp\Promise\Each::ofLimit().
If your application creates a cURL handler directly and needs deterministic
cleanup, keep a reference to the handler and call close() when the handler is
no longer needed.
use GuzzleHttp\Client;
use GuzzleHttp\Handler\CurlMultiHandler;
use GuzzleHttp\HandlerStack;
$handler = new CurlMultiHandler();
$client = new Client(['handler' => HandlerStack::create($handler)]);
try {
$client->request('GET', 'https://example.com');
} finally {
$handler->close();
}
After a cURL handler has been closed, it cannot be reused. Client and
HandlerStack do not expose close(), so applications that need deterministic
cleanup should keep the handler reference. If CurlMultiHandler::close() closes
pending transfers, their promises are rejected with
GuzzleHttp\Exception\HandlerClosedException. Explicit close() calls may
throw if native cleanup fails; destructor cleanup remains best-effort and
non-throwing.
You can pass allow-listed custom stream context options using the stream_context key of the request option. The stream_context array is an associative array where each key is a PHP transport, and each value is an associative array of transport options. Stream context options outside the built-in stream handler allow-list are rejected.
For example, let's say you need to customize the outgoing network interface used with a client.
$client->request('GET', '/', [
'stream' => true,
'stream_context' => [
'socket' => [
'bindto' => 'xxx.xxx.xxx.xxx'
]
]
]);
Custom stream context options remain active during redirects unless Guzzle
documents otherwise. See allow_redirects
for cross-origin redirect credential behavior.
You need to specify the path on disk to the CA bundle used by Guzzle for
verifying the peer certificate. See the
verify option.
Maximum function nesting level of '100' reached, aborting
You could run into this error if you have the XDebug extension installed and you execute a lot of requests in callbacks. This error message comes specifically from the XDebug extension. PHP itself does not have a function nesting limit. Change this setting in your php.ini to increase the limit:
xdebug.max_nesting_level = 1000
This can occur for a number of reasons, but if you are sending PUT, POST, or
PATCH requests with an Expect: 100-Continue header, a server that does not
support this header will return a 417 response. You can work around this by
setting the expect request option to false:
$client = new GuzzleHttp\Client();
// Disable the expect header on a single request
$response = $client->request('PUT', '/', ['expect' => false]);
// Disable the expect header on all client requests
$client = new GuzzleHttp\Client(['expect' => false]);
You can enable tracking of redirected URIs and status codes via the
track_redirects option. Each redirected URI and status code will be stored in
the X-Guzzle-Redirect-History and the X-Guzzle-Redirect-Status-History
header respectively.
The initial request's URI and the final status code will be excluded from the results. With this in mind you should be able to easily track a request's full redirect path.
For example, let's say you need to track redirects and provide both results together in a single report:
// First you configure Guzzle with redirect tracking and make a request
$client = new Client([
RequestOptions::ALLOW_REDIRECTS => [
'max' => 10, // allow at most 10 redirects.
'strict' => true, // use "strict" RFC compliant redirects.
'referer' => true, // add a Referer header
'track_redirects' => true,
],
]);
$initialRequest = '/redirect/3'; // Store the request URI for later use
$response = $client->request('GET', $initialRequest); // Make your request
// Retrieve both Redirect History headers
$redirectUriHistory = $response->getHeader('X-Guzzle-Redirect-History'); // retrieve Redirect URI history
$redirectCodeHistory = $response->getHeader('X-Guzzle-Redirect-Status-History'); // retrieve Redirect HTTP Status history
// Add the initial URI requested to the (beginning of) URI history
array_unshift($redirectUriHistory, $initialRequest);
// Add the final HTTP status code to the end of HTTP response history
array_push($redirectCodeHistory, $response->getStatusCode());
// (Optional) Combine the items of each array into a single result set
$fullRedirectReport = [];
foreach ($redirectUriHistory as $key => $value) {
$fullRedirectReport[$key] = ['location' => $value, 'code' => $redirectCodeHistory[$key]];
}
echo json_encode($fullRedirectReport);