LU-11468 lnet: set recovery interval from lnetctl

Configure lnet_recovery_interval from lnetctl

Test-Parameters: trivial
Signed-off-by: Amir Shehata <ashehata@whamcloud.com>
Change-Id: I75e3e40ea1eb87bcd6599caa4707f53cc33abea4
Reviewed-on: https://review.whamcloud.com/33498
Tested-by: Jenkins
Reviewed-by: Doug Oucharek <dougso@me.com>
Reviewed-by: Sonia Sharma <sharmaso@whamcloud.com>
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lnet/utils/lnetconfig/liblnetconfig.c b/lnet/utils/lnetconfig/liblnetconfig.c
index a3d8a5c..a194ce0 100644 (file)
--- a/lnet/utils/lnetconfig/liblnetconfig.c
+++ b/lnet/utils/lnetconfig/liblnetconfig.c
@@ -2529,6 +2529,28 @@ int ioctl_set_value(__u32 val, int ioc, char *name,
        return rc;
 }
 
+int lustre_lnet_config_recov_intrv(int intrv, int seq_no, struct cYAML **err_rc)
+{
+       int rc = LUSTRE_CFG_RC_NO_ERR;
+       char err_str[LNET_MAX_STR_LEN];
+       char val[LNET_MAX_STR_LEN];
+
+       snprintf(err_str, sizeof(err_str), "\"success\"");
+
+       snprintf(val, sizeof(val), "%d", intrv);
+
+       rc = write_sysfs_file(modparam_path, "lnet_recovery_interval", val,
+                             1, strlen(val) + 1);
+       if (rc)
+               snprintf(err_str, sizeof(err_str),
+                        "\"cannot configure recovery interval: %s\"",
+                        strerror(errno));
+
+       cYAML_build_error(rc, seq_no, ADD_CMD, "recovery_interval", err_str, err_rc);
+
+       return rc;
+}
+
 int lustre_lnet_config_hsensitivity(int sen, int seq_no, struct cYAML **err_rc)
 {
        int rc = LUSTRE_CFG_RC_NO_ERR;
@@ -3382,6 +3404,31 @@ static int ioctl_show_global_values(int ioc, int seq_no, char *name,
                                       data.sv_value, show_rc, err_rc, l_errno);
 }
 
+int lustre_lnet_show_recov_intrv(int seq_no, struct cYAML **show_rc,
+                                struct cYAML **err_rc)
+{
+       int rc = LUSTRE_CFG_RC_OUT_OF_MEM;
+       char val[LNET_MAX_STR_LEN];
+       int intrv = -1, l_errno = 0;
+       char err_str[LNET_MAX_STR_LEN];
+
+       snprintf(err_str, sizeof(err_str), "\"out of memory\"");
+
+       rc = read_sysfs_file(modparam_path, "lnet_recovery_interval", val,
+                            1, sizeof(val));
+       if (rc) {
+               l_errno = -errno;
+               snprintf(err_str, sizeof(err_str),
+                        "\"cannot get recovery interval: %d\"", rc);
+       } else {
+               intrv = atoi(val);
+       }
+
+       return build_global_yaml_entry(err_str, sizeof(err_str), seq_no,
+                                      "recovery_interval", intrv, show_rc,
+                                      err_rc, l_errno);
+}
+
 int lustre_lnet_show_hsensitivity(int seq_no, struct cYAML **show_rc,
                                  struct cYAML **err_rc)
 {
@@ -4512,7 +4559,7 @@ static int handle_yaml_config_global_settings(struct cYAML *tree,
                                              struct cYAML **err_rc)
 {
        struct cYAML *max_intf, *numa, *discovery, *retry, *tto, *seq_no,
-                    *sen;
+                    *sen, *recov;
        int rc = 0;
 
        seq_no = cYAML_get_object_item(tree, "seq_no");
@@ -4558,6 +4605,13 @@ static int handle_yaml_config_global_settings(struct cYAML *tree,
                                                        : -1,
                                                     err_rc);
 
+       recov = cYAML_get_object_item(tree, "recovery_interval");
+       if (recov)
+               rc = lustre_lnet_config_recov_intrv(recov->cy_valueint,
+                                                   seq_no ? seq_no->cy_valueint
+                                                       : -1,
+                                                   err_rc);
+
        return rc;
 }
 
@@ -4599,7 +4653,7 @@ static int handle_yaml_show_global_settings(struct cYAML *tree,
                                            struct cYAML **err_rc)
 {
        struct cYAML *max_intf, *numa, *discovery, *retry, *tto, *seq_no,
-                    *sen;
+                    *sen, *recov;
        int rc = 0;
 
        seq_no = cYAML_get_object_item(tree, "seq_no");
@@ -4639,6 +4693,12 @@ static int handle_yaml_show_global_settings(struct cYAML *tree,
                                                        : -1,
                                                     show_rc, err_rc);
 
+       recov = cYAML_get_object_item(tree, "recovery_interval");
+       if (recov)
+               rc = lustre_lnet_show_recov_intrv(seq_no ? seq_no->cy_valueint
+                                                       : -1,
+                                                 show_rc, err_rc);
+
        return rc;
 }
 

diff --git a/lnet/utils/lnetconfig/liblnetconfig.h b/lnet/utils/lnetconfig/liblnetconfig.h
index 4cd6d5e..504b797 100644 (file)
--- a/lnet/utils/lnetconfig/liblnetconfig.h
+++ b/lnet/utils/lnetconfig/liblnetconfig.h
@@ -255,6 +255,30 @@ int lustre_lnet_config_peer_ni_healthv(int value, bool all, char *pni_nid,
                                       int seq_no, struct cYAML **err_rc);
 
 /*
+ * lustre_lnet_config_recov_intrv
+ *   set the recovery interval in seconds. That's the interval to ping an
+ *   unhealthy interface.
+ *
+ *   intrv - recovery interval value to configure
+ *   seq_no - sequence number of the request
+ *   err_rc - [OUT] struct cYAML tree describing the error. Freed by
+ *   caller
+ */
+int lustre_lnet_config_recov_intrv(int intrv, int seq_no, struct cYAML **err_rc);
+
+/*
+ * lustre_lnet_show_recov_intrv
+ *    show the recovery interval set in the system
+ *
+ *   seq_no - sequence number of the request
+ *   show_rc - [OUT] struct cYAML tree containing health sensitivity info
+ *   err_rc - [OUT] struct cYAML tree describing the error. Freed by
+ *   caller
+ */
+int lustre_lnet_show_recov_intrv(int seq_no, struct cYAML **show_rc,
+                                struct cYAML **err_rc);
+
+/*
  * lustre_lnet_config_hsensitivity
  *   sets the health sensitivity; the value by which to decrement the
  *   health value of a local or peer NI. If 0 then health is turned off

diff --git a/lnet/utils/lnetctl.c b/lnet/utils/lnetctl.c
index 4932aaf..504e189 100644 (file)
--- a/lnet/utils/lnetctl.c
+++ b/lnet/utils/lnetctl.c
@@ -56,6 +56,7 @@ static int jt_set_large(int argc, char **argv);
 static int jt_set_numa(int argc, char **argv);
 static int jt_set_retry_count(int argc, char **argv);
 static int jt_set_transaction_to(int argc, char **argv);
+static int jt_set_recov_intrv(int argc, char **argv);
 static int jt_set_hsensitivity(int argc, char **argv);
 static int jt_add_peer_nid(int argc, char **argv);
 static int jt_del_peer_nid(int argc, char **argv);
@@ -200,6 +201,8 @@ command_t set_cmds[] = {
        {"health_sensitivity", jt_set_hsensitivity, 0, "sensitivity to failure\n"
         "\t0 - turn off health evaluation\n"
         "\t>0 - sensitivity value not more than 1000\n"},
+       {"recovery_interval", jt_set_recov_intrv, 0, "interval to ping in seconds (at least 1)\n"
+        "\t>0 - time in seconds between pings\n"},
        { 0, 0, 0, NULL }
 };
 
@@ -358,6 +361,34 @@ static int jt_set_numa(int argc, char **argv)
        return rc;
 }
 
+static int jt_set_recov_intrv(int argc, char **argv)
+{
+       long int value;
+       int rc;
+       struct cYAML *err_rc = NULL;
+
+       rc = check_cmd(set_cmds, "set", "recovery_interval", 2, argc, argv);
+       if (rc)
+               return rc;
+
+       rc = parse_long(argv[1], &value);
+       if (rc != 0) {
+               cYAML_build_error(-1, -1, "parser", "set",
+                                 "cannot parse recovery interval value", &err_rc);
+               cYAML_print_tree2file(stderr, err_rc);
+               cYAML_free_tree(err_rc);
+               return -1;
+       }
+
+       rc = lustre_lnet_config_recov_intrv(value, -1, &err_rc);
+       if (rc != LUSTRE_CFG_RC_NO_ERR)
+               cYAML_print_tree2file(stderr, err_rc);
+
+       cYAML_free_tree(err_rc);
+
+       return rc;
+}
+
 static int jt_set_hsensitivity(int argc, char **argv)
 {
        long int value;
@@ -1222,6 +1253,12 @@ static int jt_show_global(int argc, char **argv)
                goto out;
        }
 
+       rc = lustre_lnet_show_recov_intrv(-1, &show_rc, &err_rc);
+       if (rc != LUSTRE_CFG_RC_NO_ERR) {
+               cYAML_print_tree2file(stderr, err_rc);
+               goto out;
+       }
+
        if (show_rc)
                cYAML_print_tree(show_rc);