Drop laddr tag from tcp_stats; keep raddr for per-host breakdown

laddr is always the engine's own IP — constant, adds no diagnostic value
as a tag. Dropping it reduces row size (~15 bytes/row) and removes a tag
that was arbitrary in LocalTCPStatsCollector's default mode anyway.

raddr is kept so callers can split tcp_stats by remote host (e.g. NFS
throughput per VDB host, as Craig confirmed is needed for PerfDB).

Aggregation key changes from (laddr, raddr, service) → (raddr, service).
telegraf.base updated to remove laddr from csv_column_names and
csv_tag_columns.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: dbshah12

telegraf/connstat-stats.sh

@@ -1,17 +1,18 @@
 #!/usr/bin/env python3
 #
-# Collect per-connection TCP stats from connstat and aggregate by remote
-# endpoint (laddr:raddr:service) to bound cardinality on engines with many
-# connections — e.g. Oracle dNFS (hundreds of connections per VDB host) or
-# Elastic Data (many connections per object storage endpoint IP).
-# Mirrors the aggregation done by LocalTCPStatsCollector in the mgmt stack.
+# Collect per-connection TCP stats from connstat and aggregate by
+# (raddr, service) to mirror the behaviour of LocalTCPStatsCollector.
+#
+# laddr (local address) is intentionally excluded: it is always the engine's
+# own IP and adds no diagnostic value as a tag. raddr is kept so that callers
+# can split stats by remote host — e.g. to see NFS throughput per VDB host.
 #
 # Service name lookup reads from /etc/services, matching LocalTCPStatsCollector
 # exactly. lport is checked before rport so that listening services (where the
 # engine is the server) are identified correctly. Falls back to "unknown".
 #
-# Output fields per aggregated endpoint:
-#   laddr, raddr, service
+# Output fields per aggregated (raddr, service) group:
+#   raddr, service
 #   inbytes, outbytes, retranssegs, suna, unsent  (summed across connections)
 #   swnd, cwnd, rwnd, rtt                          (averaged across connections)
 #   connections                                    (count of aggregated conns)
@@ -69,9 +70,9 @@ for raw in proc.stdout:
     line = raw.rstrip('\n')
     if line.startswith('='):
         for key, n in cnt.items():
-            la, ra, sv = key
+            ra, sv = key
             sys.stdout.write(
-                f"{la},{ra},{sv},"
+                f"{ra},{sv},"
                 f"{inb[key]},{outb[key]},{ret_[key]},{sun[key]},{uns[key]},"
                 f"{sw[key]//n},{cw[key]//n},{rw[key]//n},{rt[key]//n},{n}\n"
             )
@@ -83,7 +84,7 @@ for raw in proc.stdout:
     fields = line.split(',')
     if len(fields) != 13:
         continue
-    la, lp, ra, rp = fields[0], fields[1], fields[2], fields[3]
+    lp, ra, rp = fields[1], fields[2], fields[3]
     lp_i = int(lp) if lp.isdigit() else 0
     rp_i = int(rp) if rp.isdigit() else 0
     if lp_i in svc:
@@ -93,7 +94,7 @@ for raw in proc.stdout:
     else:
         sv = 'unknown'
 
-    key = (la, ra, sv)
+    key = (ra, sv)
     cnt[key]  = cnt.get(key, 0)  + 1
     inb[key]  = inb.get(key, 0)  + int(fields[4])
     outb[key] = outb.get(key, 0) + int(fields[5])

telegraf/telegraf.base

@@ -59,10 +59,10 @@
   fieldpass = ["tcp*","bytes*","packets*","err*","drop*"]
 
 # Per-endpoint TCP stats (bytes, RTT, window sizes) via connstat.
-# Aggregated by remote endpoint (laddr:raddr:rport) to mirror the aggregation
-# in LocalTCPStatsCollector — avoids cardinality explosion on Oracle dNFS
-# engines (hundreds of connections per VDB host) and Elastic Data engines
-# (many connections per object storage endpoint IP).
+# Aggregated by (raddr, service): laddr (always the engine's own IP) is dropped
+# as it adds no diagnostic value. raddr is kept so stats can be split by remote
+# host — e.g. NFS throughput per VDB host. Matches LocalTCPStatsCollector's
+# default aggregation behaviour.
 # Cumulative fields (inbytes, outbytes, etc.) are summed; window/RTT fields
 # are averaged; connections = number of TCP connections aggregated.
 [[inputs.execd]]
@@ -73,9 +73,9 @@
   data_format = "csv"
   csv_delimiter = ","
   csv_trim_space = true
-  csv_column_names = ["laddr", "raddr", "service", "inbytes", "outbytes", "retranssegs", "suna", "unsent", "swnd", "cwnd", "rwnd", "rtt", "connections"]
-  csv_column_types = ["string", "string", "string", "int", "int", "int", "int", "int", "int", "int", "int", "int", "int"]
-  csv_tag_columns = ["laddr", "raddr", "service"]
+  csv_column_names = ["raddr", "service", "inbytes", "outbytes", "retranssegs", "suna", "unsent", "swnd", "cwnd", "rwnd", "rtt", "connections"]
+  csv_column_types = ["string", "string", "int", "int", "int", "int", "int", "int", "int", "int", "int", "int"]
+  csv_tag_columns = ["raddr", "service"]
 
 # Track CPU and Memory for the "delphix-mgmt" service (and children).
 [[inputs.procstat]]