Improve opentracing for incoming HTTP requests (#11618)

* remove `start_active_span_from_request`

Instead, pull out a separate function, `span_context_from_request`, to extract
the parent span, which we can then pass into `start_active_span` as
normal. This seems to be clearer all round.

* Remove redundant tags from `incoming-federation-request`

These are all wrapped up inside a parent span generated in AsyncResource, so
there's no point duplicating all the tags that are set there.

* Leave request spans open until the request completes

It may take some time for the response to be encoded into JSON, and that JSON
to be streamed back to the client, and really we want that inside the top-level
span, so let's hand responsibility for closure to the SynapseRequest.

* opentracing logs for HTTP request events

* changelog

1 files changed, 29 insertions, 1 deletions

diff --git a/synapse/http/site.py b/synapse/http/site.py
index 9f68d7e191..80f7a2ff58 100644
--- a/synapse/http/site.py
+++ b/synapse/http/site.py
@@ -14,7 +14,7 @@
 import contextlib
 import logging
 import time
-from typing import Any, Generator, Optional, Tuple, Union
+from typing import TYPE_CHECKING, Any, Generator, Optional, Tuple, Union
 
 import attr
 from zope.interface import implementer
@@ -35,6 +35,9 @@ from synapse.logging.context import (
 )
 from synapse.types import Requester
 
+if TYPE_CHECKING:
+    import opentracing
+
 logger = logging.getLogger(__name__)
 
 _next_request_seq = 0
@@ -81,6 +84,10 @@ class SynapseRequest(Request):
         # server name, for client requests this is the Requester object.
         self._requester: Optional[Union[Requester, str]] = None
 
+        # An opentracing span for this request. Will be closed when the request is
+        # completely processed.
+        self._opentracing_span: "Optional[opentracing.Span]" = None
+
         # we can't yet create the logcontext, as we don't know the method.
         self.logcontext: Optional[LoggingContext] = None
 
@@ -148,6 +155,13 @@ class SynapseRequest(Request):
         # If there's no authenticated entity, it was the requester.
         self.logcontext.request.authenticated_entity = authenticated_entity or requester
 
+    def set_opentracing_span(self, span: "opentracing.Span") -> None:
+        """attach an opentracing span to this request
+
+        Doing so will cause the span to be closed when we finish processing the request
+        """
+        self._opentracing_span = span
+
     def get_request_id(self) -> str:
         return "%s-%i" % (self.get_method(), self.request_seq)
 
@@ -286,6 +300,9 @@ class SynapseRequest(Request):
             self._processing_finished_time = time.time()
             self._is_processing = False
 
+            if self._opentracing_span:
+                self._opentracing_span.log_kv({"event": "finished processing"})
+
             # if we've already sent the response, log it now; otherwise, we wait for the
             # response to be sent.
             if self.finish_time is not None:
@@ -299,6 +316,8 @@ class SynapseRequest(Request):
         """
         self.finish_time = time.time()
         Request.finish(self)
+        if self._opentracing_span:
+            self._opentracing_span.log_kv({"event": "response sent"})
         if not self._is_processing:
             assert self.logcontext is not None
             with PreserveLoggingContext(self.logcontext):
@@ -333,6 +352,11 @@ class SynapseRequest(Request):
         with PreserveLoggingContext(self.logcontext):
             logger.info("Connection from client lost before response was sent")
 
+            if self._opentracing_span:
+                self._opentracing_span.log_kv(
+                    {"event": "client connection lost", "reason": str(reason.value)}
+                )
+
             if not self._is_processing:
                 self._finished_processing()
 
@@ -421,6 +445,10 @@ class SynapseRequest(Request):
             usage.evt_db_fetch_count,
         )
 
+        # complete the opentracing span, if any.
+        if self._opentracing_span:
+            self._opentracing_span.finish()
+
         try:
             self.request_metrics.stop(self.finish_time, self.code, self.sentLength)
         except Exception as e: