Add (admitedly redundant) clarification on kernel tasks

examples/plugins/http-requests/http-requests.c

@@ -1,7 +1,11 @@
 #include <linux/skbuff.h>
 #include <net/sock.h>
 
-/* Table from (Process id|Task id) to (Number of received http requests) */
+/* Table from (Task group id|Task id) to (Number of received http requests).
+   We need to gather requests per task and not only per task group (i.e. userspace pid)
+   so that entries can be cleared up independently when a task exists.
+   This implies that userspace needs to do the per-process aggregation.
+ */
 BPF_HASH(received_http_requests, u64, u64);
 
 

examples/plugins/http-requests/http-requests.py

@@ -27,10 +27,14 @@ def __init__(self):
         self.lock = threading.Lock()
 
     def update_http_rate_per_pid(self, last_req_count_snapshot):
-        # Aggregate per-task http request counts into per-process counts
+        # Aggregate the kernel's per-task http request counts into userland's
+        # per-process counts
         req_count_table = self.bpf.get_table(EBPF_TABLE_NAME)
         new_req_count_snapshot = collections.defaultdict(int)
         for pid_tgid, req_count in req_count_table.iteritems():
+            # Note that the kernel's tgid maps into userland's pid
+            # (not to be confused by the kernel's pid, which is
+            #  the unique identifier of a kernel task)
             pid = pid_tgid.value >> 32
             new_req_count_snapshot[pid] += req_count.value