powerman: support new --diag option

man/powerman.1.in

@@ -85,6 +85,10 @@ Retry connect to server up to N times with a 100ms delay after each failure.
 .I "-d, --device"
 Displays device status information for the device(s) that control the targets,
 if specified, or all devices if not.
+.Tp
+.I "-D, --diag"
+Display diagnostic errors for power control/query failures. Support dependent
+on device support and not available on many devices.
 .SH "TARGET SPECIFICATION"
 .B powerman
 target hostnames may be specified as comma separated or space separated

man/powerman.dev.5.in

@@ -161,6 +161,12 @@ the second argument, result in the plug state being set to off or on.
 Yes we are applying regexes to regmatches!
 If no off or on strings are provided, state will be unknown.
 .TP
+.I "setresult <regmatch> <regmatch> success=<string>"
+Set the result state. The first argument, a <regmatch> from the
+previous expect which contains the plug name. The success string is
+a compiled regex, which if it matches the second argument, results in
+the power operation being successful for this plug.
+.TP
 .I "ifoff, ifon"
 Script statements enclosed in an ifon/ifoff block are conditional
 executed based on the state of the plug passed in as an argument.

src/powerman/Makefile.am

@@ -2,6 +2,7 @@ AM_CFLAGS = @WARNING_CFLAGS@
 
 AM_CPPFLAGS = \
  -I$(top_srcdir)/src/liblsd \
+ -I$(top_srcdir)/src/libczmq \
  -I$(top_srcdir)/src/libcommon
 
 bin_PROGRAMS = \
@@ -59,6 +60,7 @@ powermand_SOURCES = \
 
 powermand_LDADD = \
  $(top_builddir)/src/liblsd/liblsd.la \
+ $(top_builddir)/src/libczmq/libczmq.la \
  $(top_builddir)/src/libcommon/libcommon.la \
  $(LIBWRAP)
 

src/powerman/arglist.h

@@ -17,11 +17,16 @@
 #define PM_ARGLIST_H
 
 typedef enum { ST_UNKNOWN, ST_OFF, ST_ON } InterpState;
+/* result not required to be set after power operation,
+ * result defaults to RT_NONE in that case.
+ */
+typedef enum { RT_NONE, RT_UNKNOWN, RT_SUCCESS } InterpResult;
 
 typedef struct {
  char *node; /* node name (in) */
  char *val; /* value as returned by the device (out) */
  InterpState state; /* interpreted value, if appropriate (out) */
+ InterpResult result; /* interpreted result, if appropriate (out) */
 } Arg;
 
 typedef struct arglist_iterator *ArgListIterator;

src/powerman/client.c

@@ -50,6 +50,7 @@
 #include "device_private.h"
 #include "fdutil.h"
 #include "powerman.h"
+#include "czmq.h"
 
 #ifndef HAVE_SOCKLEN_T
 typedef int socklen_t; /* socklen_t is uint32_t in Posix.1g */
@@ -80,6 +81,7 @@ typedef struct {
  int client_id; /* client identifier */
  bool telemetry; /* client wants telemetry debugging info */
  bool exprange; /* client wants host ranges expanded */
+ bool diag; /* client wants extra node error diagnostics */
  bool client_quit; /* set true after client quit command */
 } Client;
 
@@ -363,13 +365,99 @@ static void _client_query_device_reply(Client * c, char *arg)
  _client_printf(c, CP_RSP_QRY_COMPLETE);
 }
 
+/* zhashx_destructor_fn */
+void hostlist_destroy_wrapper(void **data)
+{
+ if (data) {
+ hostlist_t hl = *data;
+ hostlist_destroy(hl);
+ }
+}
+
+static void store_diag_msg(zhashx_t *diagmsgs, Arg *arg)
+{
+ hostlist_t hl;
+ if (!(hl = zhashx_lookup(diagmsgs, arg->val))) {
+ if (!(hl = hostlist_create(NULL)))
+ err_exit(false, "hostlist_create");
+ zhashx_insert(diagmsgs, arg->val, hl);
+ }
+ hostlist_push(hl, arg->node);
+}
+
+static void return_diags(Client *c, zhashx_t *diagmsgs)
+{
+ zlistx_t *keys = zhashx_keys(diagmsgs);
+ char *str;
+ if (!keys)
+ err_exit(false, "zhashx_keys");
+ str = zlistx_first(keys);
+ while (str) {
+ char *hosts;
+ char strbuf[1024 + 1] = {0};
+ hostlist_t hl = zhashx_lookup(diagmsgs, str);
+ assert(hl);
+ hosts = _xhostlist_ranged_string(hl);
+ snprintf(strbuf, 1024, "%s", str);
+ /* remove trailing carriage return or newline */
+ strbuf[strcspn(strbuf, "\r\n")] = '\0';
+ _client_printf(c, CP_INFO_DIAG, hosts, strbuf);
+ str = zlistx_next(keys);
+ xfree(hosts);
+ }
+ zlistx_destroy(&keys);
+}
+
+/*
+ * Reply to client power command (on/off/cycle/reset/beacon on/beacon off)
+ */
+static void _client_power_status_reply(Client * c, bool error)
+{
+ Arg *arg;
+ ArgListIterator itr;
+ int error_found = 0;
+ zhashx_t *diagmsgs = NULL;
+
+ assert(c->cmd != NULL);
+
+ /* N.B. if result is RT_NONE, device script does not
+ * specify setresult interpretation.
+ */
+ itr = arglist_iterator_create(c->cmd->arglist);
+ while ((arg = arglist_next(itr))) {
+ if (arg->result == RT_UNKNOWN) {
+ error_found++;
+ if (c->diag) {
+ if (!diagmsgs) {
+ if (!(diagmsgs = zhashx_new()))
+ err_exit(false, "zhashx_new");
+ zhashx_set_destructor(diagmsgs, hostlist_destroy_wrapper);
+ }
+ store_diag_msg(diagmsgs, arg);
+ }
+ }
+ }
+ arglist_iterator_destroy(itr);
+
+ if (c->diag && diagmsgs)
+ return_diags(c, diagmsgs);
+
+ if (c->cmd->error || error_found)
+ _client_printf(c, CP_ERR_COM_COMPLETE);
+ else
+ _client_printf(c, CP_RSP_COM_COMPLETE);
+
+ zhashx_destroy(&diagmsgs);
+}
+
 /*
  * Reply to client request for plug/soft status.
  */
 static void _client_query_status_reply(Client * c, bool error)
 {
  Arg *arg;
  ArgListIterator itr;
+ zhashx_t *diagmsgs = NULL;
 
  assert(c->cmd != NULL);
 
@@ -379,6 +467,16 @@ static void _client_query_status_reply(Client * c, bool error)
  _client_printf(c, CP_INFO_XSTATUS, arg->node,
  arg->state == ST_ON ? "on"
  : arg->state == ST_OFF ? "off" : "unknown");
+ if (arg->state == ST_UNKNOWN) {
+ if (c->diag) {
+ if (!diagmsgs) {
+ if (!(diagmsgs = zhashx_new()))
+ err_exit(false, "zhashx_new");
+ zhashx_set_destructor(diagmsgs, hostlist_destroy_wrapper);
+ }
+ store_diag_msg(diagmsgs, arg);
+ }
+ }
  }
  arglist_iterator_destroy(itr);
 
@@ -395,6 +493,14 @@ static void _client_query_status_reply(Client * c, bool error)
  switch (arg->state) {
  case ST_UNKNOWN:
  hostlist_push(hl_unknown, arg->node);
+ if (c->diag) {
+ if (!diagmsgs) {
+ if (!(diagmsgs = zhashx_new()))
+ err_exit(false, "zhashx_new");
+ zhashx_set_destructor(diagmsgs, hostlist_destroy_wrapper);
+ }
+ store_diag_msg(diagmsgs, arg);
+ }
  break;
  case ST_ON:
  hostlist_push(hl_on, arg->node);
@@ -406,6 +512,9 @@ static void _client_query_status_reply(Client * c, bool error)
  }
  arglist_iterator_destroy(itr);
 
+ if (c->diag && diagmsgs)
+ return_diags(c, diagmsgs);
+
  hostlist_sort(hl_unknown);
  hostlist_sort(hl_on);
  hostlist_sort(hl_off);
@@ -565,6 +674,9 @@ static void _parse_input(Client * c, char *input)
  } else if (!strncasecmp(str, CP_EXPRANGE, strlen(CP_EXPRANGE))) {
  c->exprange = !c->exprange; /* exprange */
  _client_printf(c, CP_RSP_EXPRANGE, c->exprange ? "ON" : "OFF");
+ } else if (!strncasecmp(str, CP_DIAG, strlen(CP_DIAG))) {
+ c->diag = !c->diag; /* diag */
+ _client_printf(c, CP_RSP_DIAG, c->diag ? "ON" : "OFF");
  } else if (!strncasecmp(str, CP_QUIT, strlen(CP_QUIT))) {
  c->client_quit = true;
  _client_printf(c, CP_RSP_QUIT); /* quit */
@@ -717,10 +829,7 @@ static void _act_finish(int client_id, ActError acterr, const char *fmt, ...)
  case PM_BEACON_OFF: /* unflash */
  case PM_POWER_CYCLE: /* cycle */
  case PM_RESET: /* reset */
- if (c->cmd->error)
- _client_printf(c, CP_ERR_COM_COMPLETE);
- else
- _client_printf(c, CP_RSP_COM_COMPLETE);
+ _client_power_status_reply(c, c->cmd->error);
  break;
  default:
  assert(false);

src/powerman/client_proto.h

@@ -48,6 +48,7 @@
 #define CP_BEACON_OFF "unflash %s"
 #define CP_TELEMETRY "telemetry"
 #define CP_EXPRANGE "exprange"
+#define CP_DIAG "diag"
 
 /*
  * Responses -
@@ -67,6 +68,7 @@
 #define CP_RSP_QRY_COMPLETE "103 Query complete" CP_EOL
 #define CP_RSP_TELEMETRY "104 Telemetry %s" CP_EOL
 #define CP_RSP_EXPRANGE "105 Hostrange expansion %s" CP_EOL
+#define CP_RSP_DIAG "106 Diag %s" CP_EOL
 
 /* failure 2xx */
 #define CP_ERR_UNKNOWN "201 Unknown command" CP_EOL
@@ -108,6 +110,7 @@
 #define CP_INFO_NODES "306 %s" CP_EOL
 #define CP_INFO_XNODES "307 %s" CP_EOL
 #define CP_INFO_ACTERROR "308 %s" CP_EOL
+#define CP_INFO_DIAG "309 %s: %s" CP_EOL
 
 #endif /* PM_CLIENT_PROTO_H */
 

src/powerman/device.c

@@ -78,6 +78,7 @@ typedef struct {
  ListIterator stmtitr; /* next stmt in block */
  Stmt *cur; /* current stmt */
  PlugListIterator plugitr; /* used by foreach */
+ PlugList pluglist; /* pluglist if foreach is ranged */
  bool processing; /* flag used by stmts, ifon/ifoff */
 } ExecCtx;
 
@@ -104,6 +105,7 @@ static bool _process_stmt(Device *dev, Action *act, ExecCtx *e,
 static bool _process_ifonoff(Device *dev, Action *act, ExecCtx *e);
 static bool _process_foreach(Device *dev, Action *act, ExecCtx *e);
 static bool _process_setplugstate(Device * dev, Action *act, ExecCtx *e);
+static bool _process_setresult(Device * dev, Action *act, ExecCtx *e);
 static bool _process_expect(Device * dev, Action *act, ExecCtx *e);
 static bool _process_send(Device * dev, Action *act, ExecCtx *e);
 static bool _process_delay(Device * dev, Action *act, ExecCtx *e,
@@ -229,6 +231,8 @@ static void _destroy_exec_ctx(ExecCtx *e)
  e->cur = NULL;
  if (e->plugs)
  list_destroy(e->plugs);
+ if (e->pluglist)
+ pluglist_destroy(e->pluglist);
  e->plugs = NULL;
  xfree(e);
 }
@@ -919,6 +923,9 @@ bool _process_stmt(Device *dev, Action *act, ExecCtx *e,
  case STMT_SETPLUGSTATE:
  finished = _process_setplugstate(dev, act, e);
  break;
+ case STMT_SETRESULT:
+ finished = _process_setresult(dev, act, e);
+ break;
  case STMT_DELAY:
  finished = _process_delay(dev, act, e, timeout);
  break;
@@ -941,8 +948,23 @@ static bool _process_foreach(Device *dev, Action *act, ExecCtx *e)
  Plug *plug = NULL;
 
  /* we store a plug iterator in the ExecCtx */
- if (e->plugitr == NULL)
- e->plugitr = pluglist_iterator_create(dev->plugs);
+ if (e->plugitr == NULL) {
+ if (act->com == PM_POWER_ON_RANGED
+ || act->com == PM_POWER_OFF_RANGED
+ || act->com == PM_POWER_CYCLE_RANGED
+ || act->com == PM_RESET_RANGED
+ || act->com == PM_BEACON_ON_RANGED
+ || act->com == PM_BEACON_OFF_RANGED) {
+ assert(e->plugs);
+ if (!e->pluglist) {
+ if (!(e->pluglist = pluglist_copy_from_list(e->plugs)))
+ goto cleanup;
+ }
+ e->plugitr = pluglist_iterator_create(e->pluglist);
+ }
+ else
+ e->plugitr = pluglist_iterator_create(dev->plugs);
+ }
 
  /* Each time the inner block is executed, its argument will be
  * a new plug name. Pick that up here.
@@ -1074,7 +1096,7 @@ static bool _process_setplugstate(Device *dev, Action *act, ExecCtx *e)
  if (str && plug && plug->node) {
  InterpState state = ST_UNKNOWN;
  ListIterator itr;
- Interp *i;
+ StateInterp *i;
  Arg *arg;
 
  itr = list_iterator_create(e->cur->u.setplugstate.interps);
@@ -1102,6 +1124,54 @@ static bool _process_setplugstate(Device *dev, Action *act, ExecCtx *e)
  return finished;
 }
 
+static bool _process_setresult(Device *dev, Action *act, ExecCtx *e)
+{
+ bool finished = true;
+ char *plug_name;
+
+ /*
+ * Usage: setresult regex regexstatus interps
+ */
+ plug_name = xregex_match_sub_strdup(dev->xmatch,
+ e->cur->u.setresult.plug_mp);
+
+ /* if no plug name, do nothing */
+ if (plug_name) {
+ char *str = xregex_match_sub_strdup(dev->xmatch,
+ e->cur->u.setresult.stat_mp);
+ Plug *plug = pluglist_find(dev->plugs, plug_name);
+
+ if (str && plug && plug->node) {
+ InterpResult result = RT_UNKNOWN;
+ ListIterator itr;
+ ResultInterp *i;
+ Arg *arg;
+
+ itr = list_iterator_create(e->cur->u.setresult.interps);
+ while ((i = list_next(itr))) {
+ if (xregex_exec(i->re, str, NULL)) {
+ result = i->result;
+ break;
+ }
+ }
+ list_iterator_destroy(itr);
+
+ if ((arg = arglist_find(act->arglist, plug->node))) {
+ arg->result = result;
+ if (arg->val)
+ xfree(arg->val);
+ arg->val = xstrdup(str);
+ }
+ }
+ if (str)
+ xfree(str);
+ /* if no match, do nothing */
+ xfree(plug_name);
+ }
+
+ return finished;
+}
+
 /* return true if expect is finished */
 static bool _process_expect(Device *dev, Action *act, ExecCtx *e)
 {