From: Yuval Lifshitz <ylifshit@ibm.com>
Date: Mon, 23 Jun 2025 10:10:37 +0000 (+0000)
Subject: rgw/notifications: add http request timeout and max inflight
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=8a2e7bfedf21180bf363a6b22fb22b0827fffc45;p=ceph.git

rgw/notifications: add http request timeout and max inflight

also make connection timeout configurable

Fixes: https://tracker.ceph.com/issues/71402

Signed-off-by: Yuval Lifshitz <ylifshit@ibm.com>
---

diff --git a/doc/radosgw/notifications.rst b/doc/radosgw/notifications.rst
index c780409f078..7cdf76e97a6 100644
--- a/doc/radosgw/notifications.rst
+++ b/doc/radosgw/notifications.rst
@@ -137,6 +137,17 @@ Notification Performance Statistics
     event on each notification, but ``pubsub_push_ok`` and ``pubsub_push_fail``
     are incremented per push action on each notification.
 
+Configuration Options
+------------------------------
+The following are global configuration options for the different endpoints:
+
+HTTP
+~~~~
+.. confval:: rgw_http_notif_message_timeout
+.. confval:: rgw_http_notif_connection_timeout
+.. confval:: rgw_http_notif_max_inflight
+
+
 Bucket Notification REST API
 ----------------------------
 
diff --git a/src/common/options/rgw.yaml.in b/src/common/options/rgw.yaml.in
index 3303ce4a6c5..607640d7276 100644
--- a/src/common/options/rgw.yaml.in
+++ b/src/common/options/rgw.yaml.in
@@ -4307,6 +4307,42 @@ options:
   services:
   - rgw
   with_legacy: true
+- name: rgw_http_notif_message_timeout
+  type: uint
+  level: advanced
+  desc: This is the maximum time in seconds to deliver a notification
+  long_desc: This is the maximum time in seconds to deliver a notification.
+    Delivery error occurs when the message timeout is exceeded.
+    This value includes the connection time, and hence must be larger than rgw_http_notif_connection_timeout.
+    If set to zero the http client will wait indefinitely.
+    see https://curl.se/libcurl/c/CURLOPT_TIMEOUT.html
+  default: 10
+  services:
+  - rgw
+  with_legacy: true
+- name: rgw_http_notif_connection_timeout
+  type: uint
+  level: advanced
+  desc: This is the maximum time in seconds to connect to an endpoint
+  long_desc: This is the maximum time in seconds to connect to an endpoint.
+    Delivery error occurs when the message timeout is exceeded.
+    If set to zero the default value of 300 seconds will be used.
+    see https://curl.se/libcurl/c/CURLOPT_CONNECTTIMEOUT.html
+  default: 5
+  services:
+  - rgw
+  with_legacy: true
+- name: rgw_http_notif_max_inflight
+  type: uint
+  level: advanced
+  desc: This is the maximum number of messages in-flight (across all http endpoints)
+  long_desc: This is the maximum number of messages in-flight (across all http endpoints).
+    Delivery error (BUSY) occurs when the number of messages is exceeded.
+    If set to zero there is no limit on the number of messages in-flight.
+  default: 8192
+  services:
+  - rgw
+  with_legacy: true
 - name: rgw_d4n_l1_datacache_address
   type: str
   level: advanced
diff --git a/src/rgw/driver/rados/rgw_pubsub_push.cc b/src/rgw/driver/rados/rgw_pubsub_push.cc
index f3baeeb0aa8..159e2b876df 100644
--- a/src/rgw/driver/rados/rgw_pubsub_push.cc
+++ b/src/rgw/driver/rados/rgw_pubsub_push.cc
@@ -61,6 +61,7 @@ bool get_bool(const RGWHTTPArgs& args, const std::string& name, bool default_val
 
 static std::unique_ptr<RGWHTTPManager> s_http_manager;
 static std::shared_mutex s_http_manager_mutex;
+static std::atomic<unsigned> s_http_manager_inflight(0);
 
 class RGWPubSubHTTPEndpoint : public RGWPubSubEndpoint {
 private:
@@ -99,10 +100,17 @@ public:
       ldout(cct, 1) << "ERROR: send failed. http endpoint manager not running" << dendl;
       return -ESRCH;
     }
+    const auto max_inflight = cct->_conf->rgw_http_notif_max_inflight;
+    if (max_inflight != 0 &&
+        s_http_manager_inflight >= max_inflight) {
+      ldout(cct, 1) << "ERROR: send failed. http endpoint manager busy. in-flight requests: " <<
+        s_http_manager_inflight << " >= " << max_inflight << dendl;
+      return -EBUSY;
+    }
     bufferlist read_bl;
     RGWPostHTTPData request(cct, "POST", endpoint, &read_bl, verify_ssl);
-    //default to 3 seconds for wrong url hits - if wrong endpoint configured
-    request.set_req_connect_timeout(3);
+    request.set_req_connect_timeout(cct->_conf->rgw_http_notif_connection_timeout);
+    request.set_req_timeout(cct->_conf->rgw_http_notif_message_timeout);
     const auto post_data = json_format_pubsub_event(event);
     if (cloudevents) {
       // following: https://github.com/cloudevents/spec/blob/v1.0.1/http-protocol-binding.md
@@ -118,11 +126,13 @@ public:
     request.set_post_data(post_data);
     request.set_send_length(post_data.length());
     request.append_header("Content-Type", "application/json");
+    ++s_http_manager_inflight;
     if (perfcounter) perfcounter->inc(l_rgw_pubsub_push_pending);
     auto rc = s_http_manager->add_request(&request);
     if (rc == 0) {
       rc = request.wait(dpp, y);
     }
+    --s_http_manager_inflight;
     if (perfcounter) perfcounter->dec(l_rgw_pubsub_push_pending);
     // TODO: use read_bl to process return code and handle according to ack level
     return rc;