public inbox for linux-nfs@vger.kernel.org
 help / color / mirror / Atom feed
From: Chuck Lever <cel@kernel.org>
To: Steve Dickson <SteveD@redhat.com>
Cc: <linux-nfs@vger.kernel.org>, Chuck Lever <chuck.lever@oracle.com>
Subject: [PATCH] mountstats: Fix per-operation percentages with nconnect
Date: Sat, 28 Feb 2026 12:46:54 -0500	[thread overview]
Message-ID: <20260228174654.129309-1-cel@kernel.org> (raw)

From: Chuck Lever <chuck.lever@oracle.com>

Per-operation percentages reported by "mountstats --rpc" are
inaccurate when an NFS mount uses nconnect.

With nconnect=N, the kernel emits N separate "xprt:" lines in
/proc/self/mountstats, one per transport.  Each transport tracks
its own rpcsends counter reflecting only RPCs routed through that
connection.

The parser overwrites rpcsends on each "xprt:" line, keeping only
the last transport's value.  Per-operation counts (READ, WRITE,
etc.) are maintained in a single array per RPC client and reflect
all RPCs across all transports.

With nconnect=3 and balanced round-robin, rpcsends holds roughly
one third of total RPCs while per-op counts hold the full total.
display_rpc_op_stats() computes (op_count * 100) / rpcsends,
yielding percentages roughly three times too large.

Accumulate rpcsends, rpcreceives, badxids, backlogutil,
sendutil, and pendutil across multiple "xprt:" lines. These are
cumulative counters where the sum across transports gives the
correct aggregate.  Per-connection properties (port, bind_count,
connect_count, connect_time, idle_time, maxslots, inflightsends)
retain the value from the last transport seen.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 tools/mountstats/mountstats.py | 61 +++++++++++++++++++++++++---------
 1 file changed, 46 insertions(+), 15 deletions(-)

diff --git a/tools/mountstats/mountstats.py b/tools/mountstats/mountstats.py
index d488f9e1c258..a6adab344d0e 100755
--- a/tools/mountstats/mountstats.py
+++ b/tools/mountstats/mountstats.py
@@ -140,6 +140,38 @@ XprtRdmaCounters = [
     'reply_waits_for_send',
 ]
 
+# Counters that should be summed across transports when nconnect > 1.
+# Each is stored in a per-transport structure in the kernel
+# (xprt->stat or rpcrdma_xprt.rx_stats) and represents a cumulative
+# event count or utilization value.  Per-connection properties (port,
+# bind_count, connect_count, connect_time, idle_time, maxslots,
+# inflightsends) retain the value from the last transport seen.
+XprtAccumulatedCounters = {
+    'rpcsends',
+    'rpcreceives',
+    'badxids',
+    'backlogutil',
+    'sendutil',
+    'pendutil',
+    'read_segments',
+    'write_segments',
+    'reply_segments',
+    'total_rdma_req',
+    'total_rdma_rep',
+    'pullup',
+    'fixup',
+    'hardway',
+    'failed_marshal',
+    'bad_reply',
+    'nomsg_calls',
+    'recovered_mrs',
+    'orphaned_mrs',
+    'allocated_mrs',
+    'local_invalidates',
+    'empty_sendctx_q',
+    'reply_waits_for_send',
+}
+
 Nfsv3ops = [
     'NULL',
     'GETATTR',
@@ -291,23 +323,22 @@ class DeviceData:
         elif words[0] == 'xprt:':
             self.__rpc_data['protocol'] = words[1]
             if words[1] == 'udp':
-                i = 2
-                for key in XprtUdpCounters:
-                    if i < len(words):
-                        self.__rpc_data[key] = int(words[i])
-                    i += 1
+                counters = XprtUdpCounters
             elif words[1] == 'tcp':
-                i = 2
-                for key in XprtTcpCounters:
-                    if i < len(words):
-                        self.__rpc_data[key] = int(words[i])
-                    i += 1
+                counters = XprtTcpCounters
             elif words[1] == 'rdma':
-                i = 2
-                for key in XprtRdmaCounters:
-                    if i < len(words):
-                        self.__rpc_data[key] = int(words[i])
-                    i += 1
+                counters = XprtRdmaCounters
+            else:
+                counters = []
+            i = 2
+            for key in counters:
+                if i < len(words):
+                    val = int(words[i])
+                    if key in XprtAccumulatedCounters and key in self.__rpc_data:
+                        self.__rpc_data[key] += val
+                    else:
+                        self.__rpc_data[key] = val
+                i += 1
         elif words[0] == 'per-op':
             self.__rpc_data['per-op'] = words
         else:
-- 
2.53.0


             reply	other threads:[~2026-02-28 17:46 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-02-28 17:46 Chuck Lever [this message]
2026-03-06 22:11 ` [PATCH] mountstats: Fix per-operation percentages with nconnect Steve Dickson

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260228174654.129309-1-cel@kernel.org \
    --to=cel@kernel.org \
    --cc=SteveD@redhat.com \
    --cc=chuck.lever@oracle.com \
    --cc=linux-nfs@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox