OPTION(mon_timecheck_interval, OPT_FLOAT, 300.0) // on leader, timecheck (clock drift check) interval (seconds)
OPTION(mon_timecheck_skew_interval, OPT_FLOAT, 30.0) // on leader, timecheck (clock drift check) interval when in presence of a skew (seconds)
OPTION(mon_pg_stuck_threshold, OPT_INT, 300) // number of seconds after which pgs can be considered inactive, unclean, or stale (see doc/control.rst under dump_stuck for more info)
-OPTION(mon_health_max_detail, OPT_INT, 50) // max detailed pgs to report in health detail
OPTION(mon_pg_min_inactive, OPT_U64, 1) // the number of PGs which have to be inactive longer than 'mon_pg_stuck_threshold' before health goes into ERR. 0 means disabled, never go into ERR.
OPTION(mon_pg_warn_min_per_osd, OPT_INT, 30) // min # pgs per (in) osd before we warn the admin
OPTION(mon_pg_warn_max_per_osd, OPT_INT, 300) // max # pgs per (in) osd before we warn the admin
OPTION(mon_health_to_clog, OPT_BOOL, true)
OPTION(mon_health_to_clog_interval, OPT_INT, 3600)
OPTION(mon_health_to_clog_tick_interval, OPT_DOUBLE, 60.0)
+OPTION(mon_health_preluminous_compat, OPT_BOOL, false)
+OPTION(mon_health_max_detail, OPT_INT, 50) // max detailed pgs to report in health detail
OPTION(mon_data_avail_crit, OPT_INT, 5)
OPTION(mon_data_avail_warn, OPT_INT, 30)
OPTION(mon_data_size_warn, OPT_U64, 15*1024*1024*1024) // issue a warning when the monitor's data store goes over 15GB (in bytes)
--- /dev/null
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_MMON_HEALTH_CHECKS_H
+#define CEPH_MMON_HEALTH_CHECKS_H
+
+#include "messages/PaxosServiceMessage.h"
+#include "mon/health_check.h"
+
+struct MMonHealthChecks : public PaxosServiceMessage
+{
+ static const int HEAD_VERSION = 1;
+ static const int COMPAT_VERSION = 1;
+
+ health_check_map_t health_checks;
+
+ MMonHealthChecks()
+ : PaxosServiceMessage(MSG_MON_HEALTH_CHECKS, HEAD_VERSION, COMPAT_VERSION) {
+ }
+ MMonHealthChecks(health_check_map_t& m)
+ : PaxosServiceMessage(MSG_MON_HEALTH_CHECKS, HEAD_VERSION, COMPAT_VERSION),
+ health_checks(m) {
+ }
+
+private:
+ ~MMonHealthChecks() override { }
+
+public:
+ const char *get_type_name() const override { return "mon_health_checks"; }
+ void print(ostream &o) const override {
+ o << "mon_health_checks(" << health_checks.checks.size() << " checks)";
+ }
+
+ void decode_payload() override {
+ bufferlist::iterator p = payload.begin();
+ paxos_decode(p);
+ ::decode(health_checks, p);
+ }
+
+ void encode_payload(uint64_t features) override {
+ paxos_encode();
+ ::encode(health_checks, payload);
+ }
+
+};
+
+#endif
LogMonitor.cc
AuthMonitor.cc
Elector.cc
+ HealthMonitor.cc
OldHealthMonitor.cc
DataHealthService.cc
PGMonitor.cc
--- /dev/null
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2013 Inktank, Inc
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software
+ * Foundation. See file COPYING.
+ *
+ */
+
+#include <stdlib.h>
+#include <limits.h>
+#include <sstream>
+#include <boost/regex.hpp>
+
+#include "include/assert.h"
+#include "include/stringify.h"
+
+#include "mon/Monitor.h"
+#include "mon/HealthService.h"
+#include "mon/HealthMonitor.h"
+#include "mon/DataHealthService.h"
+
+#include "messages/MMonHealth.h"
+#include "messages/MMonHealthChecks.h"
+
+#include "common/Formatter.h"
+
+#define dout_subsys ceph_subsys_mon
+#undef dout_prefix
+#define dout_prefix _prefix(_dout, mon, this)
+static ostream& _prefix(std::ostream *_dout, const Monitor *mon,
+ const HealthMonitor *hmon) {
+ return *_dout << "mon." << mon->name << "@" << mon->rank
+ << "(" << mon->get_state_name() << ").health ";
+}
+
+HealthMonitor::HealthMonitor(Monitor *m, Paxos *p, const string& service_name)
+ : PaxosService(m, p, service_name) {
+}
+
+void HealthMonitor::init()
+{
+ dout(10) << __func__ << dendl;
+}
+
+void HealthMonitor::create_initial()
+{
+ dout(10) << __func__ << dendl;
+}
+
+void HealthMonitor::update_from_paxos(bool *need_bootstrap)
+{
+ version = get_last_committed();
+ dout(10) << __func__ << dendl;
+ load_health();
+
+ bufferlist qbl;
+ mon->store->get(service_name, "quorum", qbl);
+ if (qbl.length()) {
+ auto p = qbl.begin();
+ ::decode(quorum_checks, p);
+ } else {
+ quorum_checks.clear();
+ }
+
+ bufferlist lbl;
+ mon->store->get(service_name, "leader", lbl);
+ if (lbl.length()) {
+ auto p = lbl.begin();
+ ::decode(leader_checks, p);
+ } else {
+ leader_checks.clear();
+ }
+
+ dout(20) << "dump:";
+ JSONFormatter jf(true);
+ jf.open_object_section("health");
+ jf.open_object_section("quorum_health");
+ for (auto& p : quorum_checks) {
+ string s = string("mon.") + stringify(p.first);
+ jf.dump_object(s.c_str(), p.second);
+ }
+ jf.close_section();
+ jf.dump_object("leader_health", leader_checks);
+ jf.close_section();
+ jf.flush(*_dout);
+ *_dout << dendl;
+}
+
+void HealthMonitor::create_pending()
+{
+ dout(10) << " " << version << dendl;
+}
+
+void HealthMonitor::encode_pending(MonitorDBStore::TransactionRef t)
+{
+ ++version;
+ dout(10) << " " << version << dendl;
+ put_last_committed(t, version);
+
+ bufferlist qbl;
+ ::encode(quorum_checks, qbl);
+ t->put(service_name, "quorum", qbl);
+ bufferlist lbl;
+ ::encode(leader_checks, lbl);
+ t->put(service_name, "leader", lbl);
+
+ health_check_map_t pending_health;
+
+ // combine per-mon details carefully...
+ map<string,set<string>> names; // code -> <mon names>
+ for (auto p : quorum_checks) {
+ for (auto q : p.second.checks) {
+ names[q.first].insert(mon->monmap->get_name(p.first));
+ }
+ pending_health.merge(p.second);
+ }
+ for (auto p : pending_health.checks) {
+ p.second.summary = boost::regex_replace(
+ p.second.summary,
+ boost::regex("%num%"), stringify(names[p.first].size()));
+ p.second.summary = boost::regex_replace(
+ p.second.summary,
+ boost::regex("%names%"), stringify(names[p.first]));
+ p.second.summary = boost::regex_replace(
+ p.second.summary,
+ boost::regex("%plurals%"),
+ names[p.first].size() > 1 ? "s" : "");
+ p.second.summary = boost::regex_replace(
+ p.second.summary,
+ boost::regex("%isorare%"),
+ names[p.first].size() > 1 ? "are" : "is");
+ }
+
+ pending_health.merge(leader_checks);
+ encode_health(pending_health, t);
+}
+
+version_t HealthMonitor::get_trim_to()
+{
+ // we don't actually need *any* old states, but keep a few.
+ if (version > 5) {
+ return version - 5;
+ }
+ return 0;
+}
+
+bool HealthMonitor::preprocess_query(MonOpRequestRef op)
+{
+ switch (op->get_req()->get_type()) {
+ case MSG_MON_HEALTH:
+ {
+ MMonHealth *hm = static_cast<MMonHealth*>(op->get_req());
+ int service_type = hm->get_service_type();
+ if (services.count(service_type) == 0) {
+ dout(1) << __func__ << " service type " << service_type
+ << " not registered -- drop message!" << dendl;
+ return false;
+ }
+ return services[service_type]->service_dispatch(op);
+ }
+
+ case MSG_MON_HEALTH_CHECKS:
+ return preprocess_health_checks(op);
+ }
+ return false;
+}
+
+bool HealthMonitor::prepare_update(MonOpRequestRef op)
+{
+ return false;
+}
+
+bool HealthMonitor::preprocess_health_checks(MonOpRequestRef op)
+{
+ MMonHealthChecks *m = static_cast<MMonHealthChecks*>(op->get_req());
+ quorum_checks[m->get_source().num()] = m->health_checks;
+ return true;
+}
+
+void HealthMonitor::tick()
+{
+ if (!is_active()) {
+ return;
+ }
+ dout(10) << __func__ << dendl;
+ bool changed = false;
+ if (check_member_health()) {
+ changed = true;
+ }
+ if (mon->is_leader()) {
+ if (check_leader_health()) {
+ changed = true;
+ }
+ }
+ if (changed) {
+ propose_pending();
+ }
+}
+
+bool HealthMonitor::check_member_health()
+{
+ dout(20) << __func__ << dendl;
+ bool changed = false;
+
+ // snapshot of usage
+ DataStats stats;
+ get_fs_stats(stats.fs_stats, g_conf->mon_data.c_str());
+ map<string,uint64_t> extra;
+ uint64_t store_size = mon->store->get_estimated_size(extra);
+ assert(store_size > 0);
+ stats.store_stats.bytes_total = store_size;
+ stats.store_stats.bytes_sst = extra["sst"];
+ stats.store_stats.bytes_log = extra["log"];
+ stats.store_stats.bytes_misc = extra["misc"];
+ stats.last_update = ceph_clock_now();
+ dout(10) << __func__ << " avail " << stats.fs_stats.avail_percent << "%"
+ << " total " << prettybyte_t(stats.fs_stats.byte_total)
+ << ", used " << prettybyte_t(stats.fs_stats.byte_used)
+ << ", avail " << prettybyte_t(stats.fs_stats.byte_avail) << dendl;
+
+ // MON_DISK_{LOW,CRIT,BIG}
+ health_check_map_t next;
+ if (stats.fs_stats.avail_percent <= g_conf->mon_data_avail_crit) {
+ stringstream ss, ss2;
+ ss << "mon%plurals% %names% %isorare% very low on available space";
+ auto& d = next.add("MON_DISK_CRIT", HEALTH_ERR, ss.str());
+ ss2 << "mon." << mon->name << " has " << stats.fs_stats.avail_percent
+ << "% avail";
+ d.detail.push_back(ss2.str());
+ } else if (stats.fs_stats.avail_percent <= g_conf->mon_data_avail_warn) {
+ stringstream ss, ss2;
+ ss << "mon%plurals% %names% %isorare% low on available space";
+ auto& d = next.add("MON_DISK_LOW", HEALTH_ERR, ss.str());
+ ss2 << "mon." << mon->name << " has " << stats.fs_stats.avail_percent
+ << "% avail";
+ d.detail.push_back(ss2.str());
+ }
+ if (stats.store_stats.bytes_total >= g_conf->mon_data_size_warn) {
+ stringstream ss, ss2;
+ ss << "mon%plurals% %names% %isorare% using a lot of disk space";
+ auto& d = next.add("MON_DISK_BIG", HEALTH_WARN, ss.str());
+ ss2 << "mon." << mon->name << " is "
+ << prettybyte_t(stats.store_stats.bytes_total)
+ << " >= mon_data_size_warn ("
+ << prettybyte_t(g_conf->mon_data_size_warn) << ")";
+ d.detail.push_back(ss2.str());
+ }
+
+ auto p = quorum_checks.find(mon->rank);
+ if (p == quorum_checks.end() ||
+ p->second != next) {
+ if (mon->is_leader()) {
+ // prepare to propose
+ quorum_checks[mon->rank] = next;
+ changed = true;
+ } else {
+ // tell the leader
+ mon->messenger->send_message(new MMonHealthChecks(next),
+ mon->monmap->get_inst(mon->get_leader()));
+ }
+ }
+ return changed;
+}
+
+bool HealthMonitor::check_leader_health()
+{
+ dout(20) << __func__ << dendl;
+ bool changed = false;
+
+ // prune quorum_health
+ {
+ auto& qset = mon->get_quorum();
+ auto p = quorum_checks.begin();
+ while (p != quorum_checks.end()) {
+ if (qset.count(p->first) == 0) {
+ p = quorum_checks.erase(p);
+ changed = true;
+ } else {
+ ++p;
+ }
+ }
+ }
+
+ health_check_map_t next;
+
+ // MON_DOWN
+ {
+ int max = mon->monmap->size();
+ int actual = mon->get_quorum().size();
+ if (actual < max) {
+ ostringstream ss;
+ ss << (max-actual) << "/" << max << " mons down, quorum "
+ << mon->get_quorum_names();
+ auto& d = next.add("MON_DOWN", HEALTH_WARN, ss.str());
+ set<int> q = mon->get_quorum();
+ for (int i=0; i<max; i++) {
+ if (q.count(i) == 0) {
+ ostringstream ss;
+ ss << "mon." << mon->monmap->get_name(i) << " (rank " << i
+ << ") addr " << mon->monmap->get_addr(i)
+ << " is down (out of quorum)";
+ d.detail.push_back(ss.str());
+ }
+ }
+ }
+ }
+
+ // MON_CLOCK_SKEW
+ if (!mon->timecheck_skews.empty()) {
+ list<string> warns;
+ list<string> details;
+ for (map<entity_inst_t,double>::iterator i = mon->timecheck_skews.begin();
+ i != mon->timecheck_skews.end(); ++i) {
+ entity_inst_t inst = i->first;
+ double skew = i->second;
+ double latency = mon->timecheck_latencies[inst];
+ string name = mon->monmap->get_name(inst.addr);
+ ostringstream tcss;
+ health_status_t tcstatus = mon->timecheck_status(tcss, skew, latency);
+ if (tcstatus != HEALTH_OK) {
+ warns.push_back(name);
+ ostringstream tmp_ss;
+ tmp_ss << "mon." << name
+ << " addr " << inst.addr << " " << tcss.str()
+ << " (latency " << latency << "s)";
+ details.push_back(tmp_ss.str());
+ }
+ }
+ if (!warns.empty()) {
+ ostringstream ss;
+ ss << "clock skew detected on";
+ while (!warns.empty()) {
+ ss << " mon." << warns.front();
+ warns.pop_front();
+ if (!warns.empty())
+ ss << ",";
+ }
+ auto& d = next.add("MON_CLOCK_SKEW", HEALTH_WARN,
+ "monitor clock skew detected");
+ d.detail.swap(details);
+ }
+ }
+
+ if (next != leader_checks) {
+ changed = true;
+ leader_checks = next;
+ }
+ return changed;
+}
--- /dev/null
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2013 Inktank, Inc
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software
+ * Foundation. See file COPYING.
+ *
+ */
+#ifndef CEPH_HEALTH_MONITOR_H
+#define CEPH_HEALTH_MONITOR_H
+
+#include "mon/PaxosService.h"
+
+//forward declaration
+namespace ceph { class Formatter; }
+class HealthService;
+
+class HealthMonitor : public PaxosService
+{
+ map<int,HealthService*> services;
+ version_t version = 0;
+ map<int,health_check_map_t> quorum_checks; // for each quorum member
+ health_check_map_t leader_checks; // leader only
+
+public:
+ HealthMonitor(Monitor *m, Paxos *p, const string& service_name);
+ ~HealthMonitor() override {
+ assert(services.empty());
+ }
+
+ /**
+ * @defgroup HealthMonitor_Inherited_h Inherited abstract methods
+ * @{
+ */
+ void init() override;
+
+ void get_health(
+ list<pair<health_status_t,string> >& summary,
+ list<pair<health_status_t,string> > *detail,
+ CephContext *cct) const override {}
+
+ bool preprocess_query(MonOpRequestRef op) override;
+ bool prepare_update(MonOpRequestRef op) override;
+
+ bool preprocess_health_checks(MonOpRequestRef op);
+ bool prepare_health_checks(MonOpRequestRef op);
+
+ bool check_leader_health();
+ bool check_member_health();
+
+ void create_initial() override;
+ void update_from_paxos(bool *need_bootstrap) override;
+ void create_pending() override;
+ void encode_pending(MonitorDBStore::TransactionRef t) override;
+ version_t get_trim_to() override;
+
+ void encode_full(MonitorDBStore::TransactionRef t) override { }
+
+ void tick() override;
+
+ /**
+ * @} // HealthMonitor_Inherited_h
+ */
+};
+
+#endif // CEPH_HEALTH_MONITOR_H
dout(4) << "active server: " << map.active_addr
<< "(" << map.active_gid << ")" << dendl;
+ load_health();
+
if (map.available) {
first_seen_inactive = utime_t();
} else {
pending_map.encode(bl, mon->get_quorum_con_features());
put_version(t, pending_map.epoch, bl);
put_last_committed(t, pending_map.epoch);
+
+ health_check_map_t next;
+ if (!pending_map.available) {
+ health_status_t level = HEALTH_WARN;
+ utime_t now = ceph_clock_now();
+ if (first_seen_inactive != utime_t() &&
+ now - first_seen_inactive > g_conf->mon_mgr_inactive_grace) {
+ level = HEALTH_ERR;
+ }
+ next.add("MGR_DOWN", level, "no active mgr");
+ }
+ encode_health(next, t);
}
bool MgrMonitor::check_caps(MonOpRequestRef op, const uuid_d& fsid)
void MgrStatMonitor::create_initial()
{
- dout(10) << dendl;
+ dout(10) << __func__ << dendl;
version = 0;
service_map.epoch = 1;
::encode(service_map, pending_service_map_bl, CEPH_FEATURES_ALL);
#include "MgrStatMonitor.h"
#include "mon/QuorumService.h"
#include "mon/OldHealthMonitor.h"
+#include "mon/HealthMonitor.h"
#include "mon/ConfigKeyService.h"
#include "common/config.h"
#include "common/cmdparse.h"
paxos_service[PAXOS_AUTH] = new AuthMonitor(this, paxos, "auth");
paxos_service[PAXOS_MGR] = new MgrMonitor(this, paxos, "mgr");
paxos_service[PAXOS_MGRSTAT] = new MgrStatMonitor(this, paxos, "mgrstat");
+ paxos_service[PAXOS_HEALTH] = new HealthMonitor(this, paxos, "health");
health_monitor = new OldHealthMonitor(this);
config_key_service = new ConfigKeyService(this, paxos);
health_status_cache.summary = summary;
}
+health_status_t Monitor::get_health_status(
+ bool want_detail,
+ Formatter *f,
+ std::string *plain,
+ const char *sep1,
+ const char *sep2)
+{
+ health_status_t r = HEALTH_OK;
+ bool compat = g_conf->mon_health_preluminous_compat;
+ if (f) {
+ f->open_object_section("health");
+ f->open_object_section("checks");
+ }
+
+ string summary;
+ string *psummary = f ? nullptr : &summary;
+ for (auto& svc : paxos_service) {
+ r = std::min(r, svc->get_health_checks().dump_summary(
+ f, psummary, sep2, want_detail));
+ }
+
+ if (f) {
+ f->close_section();
+ f->dump_stream("status") << r;
+ } else {
+ // one-liner: HEALTH_FOO[ thing1[; thing2 ...]]
+ *plain = stringify(r);
+ if (summary.size()) {
+ *plain += sep1;
+ *plain += summary;
+ }
+ *plain += "\n";
+ }
+
+ if (f && compat) {
+ f->open_array_section("summary");
+ for (auto& svc : paxos_service) {
+ svc->get_health_checks().dump_summary_compat(f);
+ }
+ f->close_section();
+ f->dump_stream("overall_status") << r;
+ }
+
+ if (want_detail) {
+ if (f && compat) {
+ f->open_array_section("detail");
+ }
+
+ for (auto& svc : paxos_service) {
+ svc->get_health_checks().dump_detail(f, plain, compat);
+ }
+
+ if (f && compat) {
+ f->close_section();
+ }
+ }
+ if (f) {
+ f->close_section();
+ }
+ return r;
+}
+
+void Monitor::log_health(
+ const health_check_map_t& updated,
+ const health_check_map_t& previous,
+ MonitorDBStore::TransactionRef t)
+{
+ if (!g_conf->mon_health_to_clog) {
+ return;
+ }
+ // FIXME: log atomically as part of @t instead of using clog.
+ dout(10) << __func__ << " updated " << updated.checks.size()
+ << " previous " << previous.checks.size()
+ << dendl;
+ for (auto& p : updated.checks) {
+ auto q = previous.checks.find(p.first);
+ if (q == previous.checks.end()) {
+ // new
+ ostringstream ss;
+ ss << p.second.severity << " " << p.first << ": "
+ << p.second.summary;
+ if (p.second.severity == HEALTH_WARN)
+ clog->warn() << ss.str();
+ else
+ clog->error() << ss.str();
+ } else {
+ if (p.second.summary != q->second.summary ||
+ p.second.severity != q->second.severity) {
+ // summary or severity changed (ignore detail changes at this level)
+ ostringstream ss;
+ ss << p.second.severity << " " << p.first << " (update): "
+ << p.second.summary;
+ if (p.second.severity == HEALTH_WARN)
+ clog->warn() << ss.str();
+ else
+ clog->error() << ss.str();
+ }
+ }
+ }
+ for (auto& p : previous.checks) {
+ if (!updated.checks.count(p.first)) {
+ // cleared
+ ostringstream ss;
+ ss << HEALTH_OK << " " << p.first << ": " << p.second.summary;
+ clog->info() << ss.str();
+ }
+ }
+}
+
health_status_t Monitor::get_health(list<string>& status,
bufferlist *detailbl,
Formatter *f)
if (f)
f->open_object_section("status");
- // reply with the status for all the components
- list<string> health;
- get_health(health, NULL, f);
-
if (f) {
f->dump_stream("fsid") << monmap->get_fsid();
+ get_health_status(false, f, nullptr);
f->dump_unsigned("election_epoch", get_epoch());
{
f->open_array_section("quorum");
f->open_object_section("fsmap");
mdsmon()->get_fsmap().print_summary(f, NULL);
f->close_section();
-
f->open_object_section("mgrmap");
mgrmon()->get_map().print_summary(f, nullptr);
f->close_section();
f->dump_object("servicemap", mgrstatmon()->get_service_map());
f->close_section();
} else {
-
ss << " cluster:\n";
ss << " id: " << monmap->get_fsid() << "\n";
- ss << " health: " << joinify(health.begin(), health.end(),
- string("\n ")) << "\n";
+
+ string health;
+ if (osdmon()->osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS) {
+ get_health_status(false, nullptr, &health,
+ "\n ", "\n ");
+ } else {
+ list<string> ls;
+ get_health(ls, NULL, f);
+ health = joinify(ls.begin(), ls.end(),
+ string("\n "));
+ }
+ ss << " health: " << health << "\n";
+
ss << "\n \n services:\n";
{
size_t maxlen = 3;
}
rdata.append(ds);
} else if (prefix == "health") {
- list<string> health_str;
- get_health(health_str, detail == "detail" ? &rdata : NULL, f.get());
- if (f) {
- f->flush(ds);
- ds << '\n';
+ if (osdmon()->osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS) {
+ string plain;
+ get_health_status(detail == "detail", f.get(), f ? nullptr : &plain);
+ if (f) {
+ f->flush(rdata);
+ } else {
+ rdata.append(plain);
+ }
} else {
- assert(!health_str.empty());
- ds << health_str.front();
- health_str.pop_front();
- if (!health_str.empty()) {
- ds << ' ';
- ds << joinify(health_str.begin(), health_str.end(), string("; "));
+ list<string> health_str;
+ get_health(health_str, detail == "detail" ? &rdata : NULL, f.get());
+ if (f) {
+ f->flush(ds);
+ ds << '\n';
+ } else {
+ assert(!health_str.empty());
+ ds << health_str.front();
+ health_str.pop_front();
+ if (!health_str.empty()) {
+ ds << ' ';
+ ds << joinify(health_str.begin(), health_str.end(), string("; "));
+ }
}
+ bufferlist comb;
+ comb.append(ds);
+ if (detail == "detail")
+ comb.append(rdata);
+ rdata = comb;
}
- bufferlist comb;
- comb.append(ds);
- if (detail == "detail")
- comb.append(rdata);
- rdata = comb;
} else if (prefix == "df") {
bool verbose = (detail == "detail");
if (f)
health_monitor->dispatch(op);
break;
+ case MSG_MON_HEALTH_CHECKS:
+ op->set_type_service();
+ paxos_service[PAXOS_HEALTH]->dispatch(op);
+ break;
+
default:
dealt_with = false;
break;
#include "common/Timer.h"
+#include "health_check.h"
#include "MonMap.h"
#include "Elector.h"
#include "Paxos.h"
version_t timecheck_round;
unsigned int timecheck_acks;
utime_t timecheck_round_start;
+ friend class HealthMonitor;
/* When we hit a skew we will start a new round based off of
* 'mon_timecheck_skew_interval'. Each new round will be backed off
* until we hit 'mon_timecheck_interval' -- which is the typical
return (class MgrStatMonitor*) paxos_service[PAXOS_MGRSTAT];
}
+ class MgrStatMonitor *healthmon() {
+ return (class MgrStatMonitor*) paxos_service[PAXOS_MGRSTAT];
+ }
+
friend class Paxos;
friend class OSDMonitor;
friend class MDSMonitor;
*/
health_status_t get_health(list<string>& status, bufferlist *detailbl,
Formatter *f);
+
+ health_status_t get_health_status(
+ bool want_detail,
+ Formatter *f,
+ std::string *plain,
+ const char *sep1 = " ",
+ const char *sep2 = "; ");
+ void log_health(
+ const health_check_map_t& updated,
+ const health_check_map_t& previous,
+ MonitorDBStore::TransactionRef t);
+
void get_cluster_status(stringstream &ss, Formatter *f);
void reply_command(MonOpRequestRef op, int rc, const string &rs, version_t version);
mapping_job.reset();
}
+ load_health();
+
/*
* We will possibly have a stashed latest that *we* wrote, and we will
* always be sure to have the oldest full map in the first..last range
::encode(pending_creatings, creatings_bl);
t->put(OSD_PG_CREATING_PREFIX, "creating", creatings_bl);
}
+
+ // health
+ _check_health(tmp, t);
+}
+
+void OSDMonitor::_check_health(
+ const OSDMap& nextmap,
+ MonitorDBStore::TransactionRef t)
+{
+ dout(20) << __func__ << dendl;
+ health_check_map_t next;
+#warning write me
+ encode_health(next, t);
}
void OSDMonitor::trim_creating_pgs(creating_pgs_t* creating_pgs,
FAST_READ_DEFAULT
};
+ void _check_health(const OSDMap& next, MonitorDBStore::TransactionRef t);
+
// svc
public:
void create_initial() override;
}
}
+void PaxosService::load_health()
+{
+ bufferlist bl;
+ mon->store->get("health", service_name, bl);
+ if (bl.length()) {
+ auto p = bl.begin();
+ ::decode(health_checks, p);
+ }
+}
*/
bool have_pending;
-protected:
+ /**
+ * health checks for this service
+ *
+ * Child must populate this during encode_pending() by calling encode_health().
+ */
+ health_check_map_t health_checks;
+public:
+ const health_check_map_t& get_health_checks() {
+ return health_checks;
+ }
+protected:
/**
* format of our state in leveldb, 0 for default
*/
version_t format_version;
-
-
/**
* @defgroup PaxosService_h_callbacks Callback classes
* @{
list<pair<health_status_t,string> > *detail,
CephContext *cct) const { }
+ void encode_health(const health_check_map_t& next,
+ MonitorDBStore::TransactionRef t) {
+ bufferlist bl;
+ ::encode(next, bl);
+ t->put("health", service_name, bl);
+ mon->log_health(next, health_checks, t);
+ }
+ void load_health();
+
private:
/**
* @defgroup PaxosService_h_store_keys Set of keys that are usually used on
--- /dev/null
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#pragma once
+
+#include <string>
+#include <map>
+
+#include "include/health.h"
+#include "common/Formatter.h"
+
+struct health_check_t {
+ health_status_t severity;
+ std::string summary;
+ std::list<std::string> detail;
+
+ DENC(health_check_t, v, p) {
+ DENC_START(1, 1, p);
+ denc(v.severity, p);
+ denc(v.summary, p);
+ denc(v.detail, p);
+ DENC_FINISH(p);
+ }
+
+ friend bool operator==(const health_check_t& l,
+ const health_check_t& r) {
+ return l.severity == r.severity &&
+ l.summary == r.summary &&
+ l.detail == r.detail;
+ }
+ friend bool operator!=(const health_check_t& l,
+ const health_check_t& r) {
+ return !(l == r);
+ }
+
+ void dump(Formatter *f) const {
+ f->dump_stream("severity") << severity;
+ f->dump_string("summary", summary);
+ f->open_array_section("detail");
+ for (auto& p : detail) {
+ f->dump_string("item", p);
+ }
+ f->close_section();
+ }
+
+ static void generate_test_instances(list<health_check_t*>& ls) {
+ ls.push_back(new health_check_t);
+ ls.push_back(new health_check_t);
+ ls.back()->severity = HEALTH_ERR;
+ ls.back()->summary = "summarization";
+ ls.back()->detail = {"one", "two", "three"};
+ }
+};
+WRITE_CLASS_DENC(health_check_t)
+
+
+struct health_check_map_t {
+ map<std::string,health_check_t> checks;
+
+ DENC(health_check_map_t, v, p) {
+ DENC_START(1, 1, p);
+ denc(v.checks, p);
+ DENC_FINISH(p);
+ }
+
+ void dump(Formatter *f) const {
+ for (auto& p : checks) {
+ f->dump_object(p.first.c_str(), p.second);
+ }
+ }
+
+ static void generate_test_instances(list<health_check_map_t*>& ls) {
+ ls.push_back(new health_check_map_t);
+ ls.push_back(new health_check_map_t);
+ {
+ auto& d = ls.back()->add("FOO", HEALTH_WARN, "foo");
+ d.detail.push_back("a");
+ d.detail.push_back("b");
+ }
+ {
+ auto& d = ls.back()->add("BAR", HEALTH_ERR, "bar!");
+ d.detail.push_back("c");
+ d.detail.push_back("d");
+ }
+ }
+
+ void clear() {
+ checks.clear();
+ }
+ void swap(health_check_map_t& other) {
+ checks.swap(other.checks);
+ }
+
+ health_check_t& add(const std::string& code,
+ health_status_t severity,
+ const std::string& summary) {
+ assert(checks.count(code) == 0);
+ health_check_t& r = checks[code];
+ r.severity = severity;
+ r.summary = summary;
+ return r;
+ }
+
+ void merge(const health_check_map_t& o) {
+ for (auto& p : o.checks) {
+ auto q = checks.find(p.first);
+ if (q == checks.end()) {
+ // new check
+ checks[p.first] = p.second;
+ } else {
+ // merge details, and hope the summary matches!
+ q->second.detail.insert(
+ q->second.detail.end(),
+ p.second.detail.begin(),
+ p.second.detail.end());
+ }
+ }
+ }
+
+ health_status_t dump_summary(Formatter *f, std::string *plain,
+ const char *sep, bool detail) const {
+ health_status_t r = HEALTH_OK;
+ for (auto& p : checks) {
+ if (r > p.second.severity) {
+ r = p.second.severity;
+ }
+ if (f) {
+ f->open_object_section(p.first.c_str());
+ f->dump_stream("severity") << p.second.severity;
+ f->dump_string("message", p.second.summary);
+ if (detail) {
+ f->open_array_section("detail");
+ for (auto& d : p.second.detail) {
+ f->dump_string("item", d);
+ }
+ f->close_section();
+ }
+ f->close_section();
+ } else {
+ if (!plain->empty()) {
+ *plain += sep;
+ }
+ *plain += p.second.summary;
+ }
+ }
+ return r;
+ }
+
+ void dump_summary_compat(Formatter *f) const {
+ for (auto& p : checks) {
+ f->open_object_section("item");
+ f->dump_stream("severity") << p.second.severity;
+ f->dump_string("summary", p.second.summary);
+ f->close_section();
+ }
+ }
+
+ void dump_detail(Formatter *f, std::string *plain, bool compat) const {
+ for (auto& p : checks) {
+ if (f) {
+ if (compat) {
+ // this is sloppy, but the best we can do: just dump all of the
+ // individual checks' details together
+ for (auto& d : p.second.detail) {
+ f->dump_string("item", d);
+ }
+ }
+ } else {
+ if (!compat) {
+ *plain += p.first + " " + p.second.summary + "\n";
+ }
+ for (auto& d : p.second.detail) {
+ if (!compat) {
+ *plain += " ";
+ }
+ *plain += d;
+ *plain += "\n";
+ }
+ }
+ }
+ }
+
+ friend bool operator==(const health_check_map_t& l,
+ const health_check_map_t& r) {
+ return l.checks == r.checks;
+ }
+ friend bool operator!=(const health_check_map_t& l,
+ const health_check_map_t& r) {
+ return !(l == r);
+ }
+};
+WRITE_CLASS_DENC(health_check_map_t)
#define PAXOS_AUTH 5
#define PAXOS_MGR 6
#define PAXOS_MGRSTAT 7
-#define PAXOS_NUM 8
+#define PAXOS_HEALTH 8
+#define PAXOS_NUM 9
inline const char *get_paxos_name(int p) {
switch (p) {
case PAXOS_AUTH: return "auth";
case PAXOS_MGR: return "mgr";
case PAXOS_MGRSTAT: return "mgrstat";
+ case PAXOS_HEALTH: return "health";
default: ceph_abort(); return 0;
}
}
#include "messages/MMonGetVersion.h"
#include "messages/MMonGetVersionReply.h"
#include "messages/MMonHealth.h"
+#include "messages/MMonHealthChecks.h"
#include "messages/MMonMetadata.h"
#include "messages/MDataPing.h"
#include "messages/MAuth.h"
case MSG_MON_HEALTH:
m = new MMonHealth();
break;
+
+ case MSG_MON_HEALTH_CHECKS:
+ m = new MMonHealthChecks();
+ break;
+
#if defined(HAVE_XIO)
case MSG_DATA_PING:
m = new MDataPing();
// Special
#define MSG_NOP 0x607
+#define MSG_MON_HEALTH_CHECKS 0x608
+
// *** ceph-mgr <-> OSD/MDS daemons ***
#define MSG_MGR_OPEN 0x700
#define MSG_MGR_CONFIGURE 0x701
projects / ceph-ci.git / commitdiff