LU-1460 snmp: Expose lnet stats through SNMP

Add to the MIB and lustre functionality to monitor
lnet stats via SNMP

Signed-off-by: Jeremy Filizetti <jeremy.filizetti@gmail.com>
Change-Id: I9ae360d7e100af5aef34a6b645fce963376928d1
Reviewed-on: http://review.whamcloud.com/4823
Tested-by: Hudson
Tested-by: Maloo <whamcloud.maloo@gmail.com>
Reviewed-by: Doug Oucharek <doug.s.oucharek@intel.com>
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>

diff --git a/snmp/Lustre-MIB.txt b/snmp/Lustre-MIB.txt
index 3780d08..f7c16b7 100644 (file)
--- a/snmp/Lustre-MIB.txt
+++ b/snmp/Lustre-MIB.txt
@@ -43,6 +43,7 @@ metaDataClients        OBJECT IDENTIFIER ::= { lustreMIB 5 }
 lustreClients          OBJECT IDENTIFIER ::= { lustreMIB 6 }
 logicalObjectVolume    OBJECT IDENTIFIER ::= { lustreMIB 7 }
 lustreLDLM             OBJECT IDENTIFIER ::= { lustreMIB 8 }
+lnetInformation        OBJECT IDENTIFIER ::= { lustreMIB 9 }
 
 
 --============================================================================
@@ -523,7 +524,7 @@ mdcMDSServerUUID OBJECT-TYPE
     DESCRIPTION
         "The Lustre Meta data server UUID accessed by Metadata Client.
          This string also contains a state name, and possibly a DEACTIVATED flag." 
-    ::= { mdcEntry 4 }    
+    ::= { mdcEntry 4 }
 
 mdcCapacity OBJECT-TYPE
     SYNTAX Counter64
@@ -919,6 +920,99 @@ ldlmResourceCount OBJECT-TYPE
 
 --============================================================================
 --
+-- Lustre Lnet
+--
+--============================================================================
+lnetMsgsAlloc OBJECT-TYPE
+    SYNTAX Unsigned32
+    MAX-ACCESS read-only
+    STATUS current
+    DESCRIPTION
+        "LNet messages allocated."
+    ::= { lnetInformation 1 }
+
+lnetMsgsMax OBJECT-TYPE
+    SYNTAX Unsigned32
+    MAX-ACCESS read-only
+    STATUS current
+    DESCRIPTION
+        "LNet messages maximum."
+    ::= { lnetInformation 2 }
+
+lnetErrors OBJECT-TYPE
+    SYNTAX Unsigned32
+    MAX-ACCESS read-only
+    STATUS current
+    DESCRIPTION
+        "LNet errors."
+    ::= { lnetInformation 3 }
+
+lnetSendCount OBJECT-TYPE
+    SYNTAX Unsigned32
+    MAX-ACCESS read-only
+    STATUS current
+    DESCRIPTION
+        "LNet send RPCs."
+    ::= { lnetInformation 4 }
+
+lnetRecvCount OBJECT-TYPE
+    SYNTAX Unsigned32
+    MAX-ACCESS read-only
+    STATUS current
+    DESCRIPTION
+        "LNet receive RPCs."
+    ::= { lnetInformation 5 }
+
+lnetRouteCount OBJECT-TYPE
+    SYNTAX Unsigned32
+    MAX-ACCESS read-only
+    STATUS current
+    DESCRIPTION
+        "LNet routed RPCs."
+    ::= { lnetInformation 6 }
+
+lnetDropCount OBJECT-TYPE
+    SYNTAX Unsigned32
+    MAX-ACCESS read-only
+    STATUS current
+    DESCRIPTION
+        "LNet dropped RPCs."
+    ::= { lnetInformation 7 }
+
+lnetSendBytes OBJECT-TYPE
+    SYNTAX Unsigned64
+    MAX-ACCESS read-only
+    STATUS current
+    DESCRIPTION
+        "LNet send bytes."
+    ::= { lnetInformation 8 }
+
+lnetRecvBytes OBJECT-TYPE
+    SYNTAX Unsigned64
+    MAX-ACCESS read-only
+    STATUS current
+    DESCRIPTION
+        "LNet receive bytes."
+    ::= { lnetInformation 9 }
+
+lnetRouteBytes OBJECT-TYPE
+    SYNTAX Unsigned64
+    MAX-ACCESS read-only
+    STATUS current
+    DESCRIPTION
+        "LNet routed bytes."
+    ::= { lnetInformation 10 }
+
+lnetDropBytes OBJECT-TYPE
+    SYNTAX Unsigned64
+    MAX-ACCESS read-only
+    STATUS current
+    DESCRIPTION
+        "LNet dropped bytes."
+    ::= { lnetInformation 11 }
+
+--============================================================================
+--
 -- Lustre Management Traps
 --
 --============================================================================

diff --git a/snmp/lustre-snmp-util.c b/snmp/lustre-snmp-util.c
index 36845ad..93908a1 100644 (file)
--- a/snmp/lustre-snmp-util.c
+++ b/snmp/lustre-snmp-util.c
@@ -763,3 +763,12 @@ extern int mds_stats_values(char * name_value, unsigned long long * nb_sample, u
   
   return SUCCESS;
 }
+
+void convert_ull(counter64 *c64, unsigned long long ull, size_t *var_len)
+{
+        *var_len  = sizeof(c64);
+        c64->low  = (unsigned long long) (0x0ffffffff & ull);
+        ull >>= 32;
+        c64->high = (unsigned long long) (0x0ffffffff & ull);
+}
+

diff --git a/snmp/lustre-snmp-util.h b/snmp/lustre-snmp-util.h
index 1710ef5..6e46d24 100644 (file)
--- a/snmp/lustre-snmp-util.h
+++ b/snmp/lustre-snmp-util.h
@@ -112,8 +112,22 @@
 #define LDLMUNUSEDLOCKCOUNT 103
 #define LDLMRESOURCECOUNT   104
 
+#define LNETNUMBER          110
+#define LNETMSGSALLOC       111
+#define LNETMSGSMAX         112
+#define LNETERRORS          113
+#define LNETSENDCOUNT       114
+#define LNETRECVCOUNT       115
+#define LNETROUTECOUNT      116
+#define LNETDROPCOUNT       117
+#define LNETSENDBYTES       118
+#define LNETRECVBYTES       119
+#define LNETROUTEBYTES      120
+#define LNETDROPBYTES       121
+
 /* Defining the proc paths for Lustre file system */
 #define LUSTRE_PATH                 "/proc/fs/lustre/"
+#define LNET_PATH                   "/proc/sys/lnet/"
 #define OSD_PATH                    LUSTRE_PATH "obdfilter/"
 #define OSC_PATH                    LUSTRE_PATH "osc/"
 #define MDS_PATH                    LUSTRE_PATH "mds/"
@@ -215,6 +229,7 @@ unsigned char *
         const char *path,
         struct oid_table *ptable);
 
+void convert_ull(counter64 *c64, unsigned long long ull, size_t *var_len);
 int stats_values(char * filepath,char * name_value, unsigned long long * nb_sample, unsigned long long * min, unsigned long long * max, unsigned long long * sum, unsigned long long * sum_square);
 extern int mds_stats_values(char * name_value, unsigned long long * nb_sample, unsigned long long * min, unsigned long long * max, unsigned long long * sum, unsigned long long * sum_square);
 

diff --git a/snmp/lustre-snmp.c b/snmp/lustre-snmp.c
index 0473852..41070e0 100644 (file)
--- a/snmp/lustre-snmp.c
+++ b/snmp/lustre-snmp.c
@@ -45,6 +45,8 @@
 #include <pthread.h>
 #include "lustre-snmp.h"
 
+#define LNET_CHECK_INTERVAL 500
+
 /* 
  * clusterFileSystems_variables_oid:
  *   this is the top level oid that we want to register under.  This
@@ -135,7 +137,7 @@ struct variable7 clusterFileSystems_variables[] = {
   /* logicalObjectVolume 2.1.7 */
   { LOVNUMBER           , ASN_UNSIGNED  , RONLY , var_clusterFileSystems, 4, { 2,1,7,1 } },
 
-  /* logicalObjectVolume.osdTable.lovTable 2.1.2.2.1 */
+  /* logicalObjectVolume.osdTable.lovTable 2.1.7.2.1 */
   { LOVUUID             , ASN_OCTET_STR , RONLY , var_lovTable, 6, { 2,1,7,2,1,2 } },
   { LOVCOMMONNAME       , ASN_OCTET_STR , RONLY , var_lovTable, 6, { 2,1,7,2,1,3 } },
   { LOVNUMOBD           , ASN_UNSIGNED ,  RONLY , var_lovTable, 6, { 2,1,7,2,1,4 } },
@@ -158,6 +160,18 @@ struct variable7 clusterFileSystems_variables[] = {
   { LDLMUNUSEDLOCKCOUNT , ASN_UNSIGNED  , RONLY , var_ldlmTable, 6, { 2,1,8,2,1,4 } },
   { LDLMRESOURCECOUNT   , ASN_UNSIGNED  , RONLY , var_ldlmTable, 6, { 2,1,8,2,1,5 } },
 
+  /* lnetInformation 2.1.9 */
+  { LNETMSGSALLOC,  ASN_UNSIGNED,  RONLY, var_lnetInformation, 4, { 2,1,9,1 } },
+  { LNETMSGSMAX,    ASN_UNSIGNED,  RONLY, var_lnetInformation, 4, { 2,1,9,2 } },
+  { LNETERRORS,     ASN_UNSIGNED,  RONLY, var_lnetInformation, 4, { 2,1,9,3 } },
+  { LNETSENDCOUNT,  ASN_UNSIGNED,  RONLY, var_lnetInformation, 4, { 2,1,9,4 } },
+  { LNETRECVCOUNT,  ASN_UNSIGNED,  RONLY, var_lnetInformation, 4, { 2,1,9,5 } },
+  { LNETROUTECOUNT, ASN_UNSIGNED,  RONLY, var_lnetInformation, 4, { 2,1,9,6 } },
+  { LNETDROPCOUNT,  ASN_UNSIGNED,  RONLY, var_lnetInformation, 4, { 2,1,9,7 } },
+  { LNETSENDBYTES,  ASN_COUNTER64, RONLY, var_lnetInformation, 4, { 2,1,9,8 } },
+  { LNETRECVBYTES,  ASN_COUNTER64, RONLY, var_lnetInformation, 4, { 2,1,9,9 } },
+  { LNETROUTEBYTES, ASN_COUNTER64, RONLY, var_lnetInformation, 4, { 2,1,9,10 } },
+  { LNETDROPBYTES,  ASN_COUNTER64, RONLY, var_lnetInformation, 4, { 2,1,9,11 } },
 };
 
 /*****************************************************************************
@@ -528,6 +542,95 @@ var_ldlmTable(struct variable *vp,
         LDLM_PATH,ldlm_table);
 }
 
+/*****************************************************************************
+ * Function: var_lnetInformation
+ *
+ ****************************************************************************/
+unsigned char *
+var_lnetInformation(struct variable *vp,
+                    oid             *name,
+                    size_t          *length,
+                    int              exact,
+                    size_t          *var_len,
+                    WriteMethod    **write_method)
+{
+        /* variables we may use later */
+        static unsigned char      string[SPRINT_MAX_LEN];
+        static unsigned int       i[7];
+        static unsigned long long ull[4];
+        static unsigned long      next_update;
+        static counter64          c64;
+        static unsigned int       c32;
+        struct timeval            current_tv;
+        unsigned long             current;
+        char                      file_path[MAX_PATH_SIZE];
+
+        /* Update at most every LNET_STATS_INTERVAL milliseconds */
+        gettimeofday(&current_tv, NULL);
+        current = current_tv.tv_sec * 1000000 + current_tv.tv_usec;
+        if (current >= next_update) {
+                sprintf(file_path, "%s%s", LNET_PATH, "stats");
+                if (read_string(file_path, (char *) string, sizeof(string))
+                    != SUCCESS)
+                        return NULL;
+
+                sscanf((char *) string,
+                       "%u %u %u %u %u %u %u %llu %llu %llu %llu",
+                       &i[0], &i[1], &i[2], &i[3], &i[4], &i[5], &i[6],
+                       &ull[0], &ull[1], &ull[2], &ull[3]);
+
+                next_update = current + (LNET_CHECK_INTERVAL * 1000);
+        }
+
+        if (header_generic(vp, name, length, exact, var_len, write_method)
+            == MATCH_FAILED)
+                return NULL;
+
+        switch (vp->magic) {
+        case LNETMSGSALLOC:
+                *var_len = sizeof(c32);
+                c32 = i[0];
+                return (unsigned char *) &c32;
+        case LNETMSGSMAX:
+                *var_len = sizeof(c32);
+                c32 = i[1];
+                return (unsigned char *) &c32;
+        case LNETERRORS:
+                *var_len = sizeof(c32);
+                c32 = i[2];
+                return (unsigned char *) &c32;
+        case LNETSENDCOUNT:
+                *var_len = sizeof(c32);
+                c32 = i[3];
+                return (unsigned char *) &c32;
+        case LNETRECVCOUNT:
+                *var_len = sizeof(c32);
+                c32 = i[4];
+                return (unsigned char *) &c32;
+        case LNETROUTECOUNT:
+                *var_len = sizeof(c32);
+                c32 = i[5];
+                return (unsigned char *) &c32;
+        case LNETDROPCOUNT:
+                *var_len = sizeof(c32);
+                c32 = i[6];
+                return (unsigned char *) &c32;
+        case LNETSENDBYTES:
+                convert_ull(&c64, ull[0], var_len);
+                return (unsigned char *) &c64;
+        case LNETRECVBYTES:
+                convert_ull(&c64, ull[1], var_len);
+                return (unsigned char *) &c64;
+        case LNETROUTEBYTES:
+                convert_ull(&c64, ull[2], var_len);
+                return (unsigned char *) &c64;
+        case LNETDROPBYTES:
+                convert_ull(&c64, ull[3], var_len);
+                return (unsigned char *) &c64;
+        default:
+                return NULL;
+        }
+}
 
 /*****************************************************************************
  * Function: var_mdsNbSampledReq

diff --git a/snmp/lustre-snmp.h b/snmp/lustre-snmp.h
index 89eddae..27eaae3 100644 (file)
--- a/snmp/lustre-snmp.h
+++ b/snmp/lustre-snmp.h
@@ -57,6 +57,7 @@ FindVarMethod var_mdcTable;
 FindVarMethod var_cliTable;
 FindVarMethod var_ldlmTable;
 FindVarMethod var_lovTable;
+FindVarMethod var_lnetInformation;
 FindVarMethod var_mdsNbSampledReq;
 WriteMethod write_sysStatus;