#include "PyModuleRegistry.h"
#include "ActivePyModules.h"
+#include "DaemonKey.h"
#include "DaemonServer.h"
#define dout_context g_ceph_context
#define dout_subsys ceph_subsys_mgr
#undef dout_prefix
-#define dout_prefix *_dout << "mgr " << __func__ << " "
+#define dout_prefix *_dout << "mgr " << __func__ << " "
ActivePyModules::ActivePyModules(PyModuleConfig &module_config_,
std::map<std::string, std::string> store_data,
: module_config(module_config_), daemon_state(ds), cluster_state(cs),
monc(mc), clog(clog_), audit_clog(audit_clog_), objecter(objecter_),
client(client_), finisher(f),
- server(server), py_module_registry(pmr), lock("ActivePyModules")
+ cmd_finisher(g_ceph_context, "cmd_finisher", "cmdfin"),
+ server(server), py_module_registry(pmr)
{
store_cache = std::move(store_data);
+ cmd_finisher.start();
}
ActivePyModules::~ActivePyModules() = default;
f->open_array_section("services");
std::string ceph_version;
- for (const auto &i : dmc) {
- std::lock_guard l(i.second->lock);
- const auto &key = i.first;
- const std::string &str_type = key.first;
- const std::string &svc_name = key.second;
-
+ for (const auto &[key, state] : dmc) {
+ std::lock_guard l(state->lock);
// TODO: pick the highest version, and make sure that
// somewhere else (during health reporting?) we are
// indicating to the user if we see mixed versions
- auto ver_iter = i.second->metadata.find("ceph_version");
- if (ver_iter != i.second->metadata.end()) {
- ceph_version = i.second->metadata.at("ceph_version");
+ auto ver_iter = state->metadata.find("ceph_version");
+ if (ver_iter != state->metadata.end()) {
+ ceph_version = state->metadata.at("ceph_version");
}
f->open_object_section("service");
- f->dump_string("type", str_type);
- f->dump_string("id", svc_name);
+ f->dump_string("type", key.type);
+ f->dump_string("id", key.name);
f->close_section();
}
f->close_section();
const std::string &svc_type,
const std::string &svc_id)
{
- auto metadata = daemon_state.get(DaemonKey(svc_type, svc_id));
+ auto metadata = daemon_state.get(DaemonKey{svc_type, svc_id});
if (metadata == nullptr) {
derr << "Requested missing service " << svc_type << "." << svc_id << dendl;
Py_RETURN_NONE;
const std::string &svc_type,
const std::string &svc_id)
{
- auto metadata = daemon_state.get(DaemonKey(svc_type, svc_id));
+ auto metadata = daemon_state.get(DaemonKey{svc_type, svc_id});
if (metadata == nullptr) {
derr << "Requested missing service " << svc_type << "." << svc_id << dendl;
Py_RETURN_NONE;
osd_map.crush->encode(rdata, CEPH_FEATURES_SUPPORTED_DEFAULT);
});
std::string crush_text = rdata.to_str();
- return PyString_FromString(crush_text.c_str());
+ return PyUnicode_FromString(crush_text.c_str());
} else if (what.substr(0, 7) == "osd_map") {
cluster_state.with_osdmap([&f, &what, &tstate](const OSDMap &osd_map){
PyEval_RestoreThread(tstate);
}
});
return f.get();
+ } else if (what == "modified_config_options") {
+ PyEval_RestoreThread(tstate);
+ auto all_daemons = daemon_state.get_all();
+ set<string> names;
+ for (auto& [key, daemon] : all_daemons) {
+ std::lock_guard l(daemon->lock);
+ for (auto& [name, valmap] : daemon->config) {
+ names.insert(name);
+ }
+ }
+ f.open_array_section("options");
+ for (auto& name : names) {
+ f.dump_string("name", name);
+ }
+ f.close_section();
+ return f.get();
} else if (what.substr(0, 6) == "config") {
PyEval_RestoreThread(tstate);
if (what == "config_options") {
auto dmc = daemon_state.get_by_service("osd");
PyEval_RestoreThread(tstate);
- for (const auto &i : dmc) {
- std::lock_guard l(i.second->lock);
- f.open_object_section(i.first.second.c_str());
- f.dump_string("hostname", i.second->hostname);
- for (const auto &j : i.second->metadata) {
- f.dump_string(j.first.c_str(), j.second);
+ for (const auto &[key, state] : dmc) {
+ std::lock_guard l(state->lock);
+ f.open_object_section(key.name.c_str());
+ f.dump_string("hostname", state->hostname);
+ for (const auto &[name, val] : state->metadata) {
+ f.dump_string(name.c_str(), val);
+ }
+ f.close_section();
+ }
+ return f.get();
+ } else if (what == "mds_metadata") {
+ auto dmc = daemon_state.get_by_service("mds");
+ PyEval_RestoreThread(tstate);
+
+ for (const auto &[key, state] : dmc) {
+ std::lock_guard l(state->lock);
+ f.open_object_section(key.name.c_str());
+ f.dump_string("hostname", state->hostname);
+ for (const auto &[name, val] : state->metadata) {
+ f.dump_string(name.c_str(), val);
}
f.close_section();
}
cluster_state.with_pgmap(
[&f, &tstate](const PGMap &pg_map) {
PyEval_RestoreThread(tstate);
- pg_map.dump(&f);
+ pg_map.dump(&f, false);
}
);
return f.get();
} else if (what.size() > 7 &&
what.substr(0, 7) == "device ") {
string devid = what.substr(7);
- daemon_state.with_device(devid, [&f, &tstate] (const DeviceState& dev) {
- PyEval_RestoreThread(tstate);
- f.dump_object("device", dev);
- });
+ if (!daemon_state.with_device(
+ devid,
+ [&f, &tstate] (const DeviceState& dev) {
+ PyEval_RestoreThread(tstate);
+ f.dump_object("device", dev);
+ })) {
+ // device not found
+ PyEval_RestoreThread(tstate);
+ }
return f.get();
} else if (what == "io_rate") {
cluster_state.with_pgmap(
return f.get();
} else if (what == "df") {
cluster_state.with_osdmap_and_pgmap(
- [this, &f, &tstate](
+ [&f, &tstate](
const OSDMap& osd_map,
const PGMap &pg_map) {
PyEval_RestoreThread(tstate);
pg_map.dump_pool_stats_full(osd_map, nullptr, &f, true);
});
return f.get();
+ } else if (what == "pg_stats") {
+ cluster_state.with_pgmap(
+ [&f, &tstate](const PGMap &pg_map) {
+ PyEval_RestoreThread(tstate);
+ pg_map.dump_pg_stats(&f, false);
+ });
+ return f.get();
+ } else if (what == "pool_stats") {
+ cluster_state.with_pgmap(
+ [&f, &tstate](const PGMap &pg_map) {
+ PyEval_RestoreThread(tstate);
+ pg_map.dump_pool_stats(&f);
+ });
+ return f.get();
+ } else if (what == "pg_ready") {
+ PyEval_RestoreThread(tstate);
+ server.dump_pg_ready(&f);
+ return f.get();
} else if (what == "osd_stats") {
cluster_state.with_pgmap(
[&f, &tstate](const PGMap &pg_map) {
PyEval_RestoreThread(tstate);
- pg_map.dump_osd_stats(&f);
+ pg_map.dump_osd_stats(&f, false);
+ });
+ return f.get();
+ } else if (what == "osd_ping_times") {
+ cluster_state.with_pgmap(
+ [&f, &tstate](const PGMap &pg_map) {
+ PyEval_RestoreThread(tstate);
+ pg_map.dump_osd_ping_times(&f);
});
return f.get();
} else if (what == "osd_pool_stats") {
int64_t poolid = -ENOENT;
- string pool_name;
cluster_state.with_osdmap_and_pgmap([&](const OSDMap& osdmap,
const PGMap& pg_map) {
PyEval_RestoreThread(tstate);
f.close_section();
});
return f.get();
- } else if (what == "health" || what == "mon_status") {
- bufferlist json;
- if (what == "health") {
- json = cluster_state.get_health();
- } else if (what == "mon_status") {
- json = cluster_state.get_mon_status();
- } else {
- ceph_abort();
- }
-
- PyEval_RestoreThread(tstate);
- f.dump_string("json", json.to_str());
+ } else if (what == "health") {
+ cluster_state.with_health(
+ [&f, &tstate](const ceph::bufferlist &health_json) {
+ PyEval_RestoreThread(tstate);
+ f.dump_string("json", health_json.to_str());
+ });
+ return f.get();
+ } else if (what == "mon_status") {
+ cluster_state.with_mon_status(
+ [&f, &tstate](const ceph::bufferlist &mon_status_json) {
+ PyEval_RestoreThread(tstate);
+ f.dump_string("json", mon_status_json.to_str());
+ });
return f.get();
} else if (what == "mgr_map") {
cluster_state.with_mgrmap([&f, &tstate](const MgrMap &mgr_map) {
{
std::lock_guard l(lock);
- ceph_assert(modules.count(py_module->get_name()) == 0);
-
const auto name = py_module->get_name();
- modules[name].reset(new ActivePyModule(py_module, clog));
- auto active_module = modules.at(name).get();
+ auto active_module = std::make_shared<ActivePyModule>(py_module, clog);
+ pending_modules.insert(name);
// Send all python calls down a Finisher to avoid blocking
// C++ code, and avoid any potential lock cycles.
- finisher.queue(new FunctionContext([this, active_module, name](int) {
+ finisher.queue(new LambdaContext([this, active_module, name](int) {
int r = active_module->load(this);
+ std::lock_guard l(lock);
+ pending_modules.erase(name);
if (r != 0) {
derr << "Failed to run module in active mode ('" << name << "')"
<< dendl;
- std::lock_guard l(lock);
- modules.erase(name);
} else {
+ auto em = modules.emplace(name, active_module);
+ ceph_assert(em.second); // actually inserted
+
dout(4) << "Starting thread for " << name << dendl;
active_module->thread.create(active_module->get_thread_name());
}
std::lock_guard locker(lock);
// Signal modules to drop out of serve() and/or tear down resources
- for (auto &i : modules) {
- auto module = i.second.get();
- const auto& name = i.first;
-
- lock.Unlock();
+ for (auto& [name, module] : modules) {
+ lock.unlock();
dout(10) << "calling module " << name << " shutdown()" << dendl;
module->shutdown();
dout(10) << "module " << name << " shutdown() returned" << dendl;
- lock.Lock();
+ lock.lock();
}
// For modules implementing serve(), finish the threads where we
// were running that.
- for (auto &i : modules) {
- lock.Unlock();
- dout(10) << "joining module " << i.first << dendl;
- i.second->thread.join();
- dout(10) << "joined module " << i.first << dendl;
- lock.Lock();
+ for (auto& [name, module] : modules) {
+ lock.unlock();
+ dout(10) << "joining module " << name << dendl;
+ module->thread.join();
+ dout(10) << "joined module " << name << dendl;
+ lock.lock();
}
+ cmd_finisher.wait_for_empty();
+ cmd_finisher.stop();
+
modules.clear();
}
std::lock_guard l(lock);
dout(10) << __func__ << ": notify_all " << notify_type << dendl;
- for (auto& i : modules) {
- auto module = i.second.get();
+ for (auto& [name, module] : modules) {
// Send all python calls down a Finisher to avoid blocking
// C++ code, and avoid any potential lock cycles.
- finisher.queue(new FunctionContext([module, notify_type, notify_id](int r){
- module->notify(notify_type, notify_id);
+ dout(15) << "queuing notify to " << name << dendl;
+ // workaround for https://bugs.llvm.org/show_bug.cgi?id=35984
+ finisher.queue(new LambdaContext([module=module, notify_type, notify_id]
+ (int r){
+ module->notify(notify_type, notify_id);
}));
}
}
std::lock_guard l(lock);
dout(10) << __func__ << ": notify_all (clog)" << dendl;
- for (auto& i : modules) {
- auto module = i.second.get();
+ for (auto& [name, module] : modules) {
// Send all python calls down a Finisher to avoid blocking
// C++ code, and avoid any potential lock cycles.
//
// Note intentional use of non-reference lambda binding on
// log_entry: we take a copy because caller's instance is
// probably ephemeral.
- finisher.queue(new FunctionContext([module, log_entry](int r){
+ dout(15) << "queuing notify (clog) to " << name << dendl;
+ // workaround for https://bugs.llvm.org/show_bug.cgi?id=35984
+ finisher.queue(new LambdaContext([module=module, log_entry](int r){
module->notify_clog(log_entry);
}));
}
const std::string global_key = PyModule::config_prefix
+ module_name + "/" + key;
- dout(4) << __func__ << " key: " << global_key << dendl;
+ dout(20) << " key: " << global_key << dendl;
std::lock_guard lock(module_config.lock);
-
+
auto i = module_config.config.find(global_key);
if (i != module_config.config.end()) {
*val = i->second;
}
PyEval_RestoreThread(tstate);
if (prefix.size()) {
- dout(4) << __func__ << " [" << prefix << "/]" << key << " not found "
+ dout(10) << " [" << prefix << "/]" << key << " not found "
<< dendl;
} else {
- dout(4) << __func__ << " " << key << " not found " << dendl;
+ dout(10) << " " << key << " not found " << dendl;
}
Py_RETURN_NONE;
}
dout(4) << __func__ << " prefix: " << global_prefix << dendl;
PyFormatter f;
-
+
for (auto p = store_cache.lower_bound(global_prefix);
p != store_cache.end() && p->first.find(global_prefix) == 0;
++p) {
{
const std::string global_key = PyModule::config_prefix
+ module_name + "/" + key;
-
+
Command set_cmd;
{
- PyThreadState *tstate = PyEval_SaveThread();
std::lock_guard l(lock);
- PyEval_RestoreThread(tstate);
-
if (val) {
store_cache[global_key] = *val;
} else {
{
std::map<std::string, std::string> result;
std::lock_guard l(lock);
- for (const auto& i : modules) {
- const auto &module = i.second.get();
+ for (const auto& [name, module] : modules) {
std::string svc_str = module->get_uri();
if (!svc_str.empty()) {
- result[module->get_name()] = svc_str;
+ result[name] = svc_str;
}
}
PyFormatter f;
f.open_array_section(path.c_str());
- auto metadata = daemon_state.get(DaemonKey(svc_name, svc_id));
+ auto metadata = daemon_state.get(DaemonKey{svc_name, svc_id});
if (metadata) {
std::lock_guard l2(metadata->lock);
if (metadata->perf_counters.instances.count(path)) {
const auto &avg_data = counter_instance.get_data_avg();
for (const auto &datapoint : avg_data) {
f.open_array_section("datapoint");
- f.dump_unsigned("t", datapoint.t.sec());
+ f.dump_float("t", datapoint.t);
f.dump_unsigned("s", datapoint.s);
f.dump_unsigned("c", datapoint.c);
f.close_section();
const auto &data = counter_instance.get_data();
for (const auto &datapoint : data) {
f.open_array_section("datapoint");
- f.dump_unsigned("t", datapoint.t.sec());
+ f.dump_float("t", datapoint.t);
f.dump_unsigned("v", datapoint.v);
f.close_section();
}
{
if (counter_type.type & PERFCOUNTER_LONGRUNAVG) {
const auto &datapoint = counter_instance.get_latest_data_avg();
- f.dump_unsigned("t", datapoint.t.sec());
+ f.dump_float("t", datapoint.t);
f.dump_unsigned("s", datapoint.s);
f.dump_unsigned("c", datapoint.c);
} else {
const auto &datapoint = counter_instance.get_latest_data();
- f.dump_unsigned("t", datapoint.t.sec());
+ f.dump_float("t", datapoint.t);
f.dump_unsigned("v", datapoint.v);
}
};
} else if (svc_id.empty()) {
daemons = daemon_state.get_by_service(svc_type);
} else {
- auto key = DaemonKey(svc_type, svc_id);
+ auto key = DaemonKey{svc_type, svc_id};
// so that the below can be a loop in all cases
auto got = daemon_state.get(key);
if (got != nullptr) {
PyFormatter f;
if (!daemons.empty()) {
- for (auto statepair : daemons) {
- auto key = statepair.first;
- auto state = statepair.second;
-
- std::ostringstream daemon_name;
- daemon_name << key.first << "." << key.second;
- f.open_object_section(daemon_name.str().c_str());
+ for (auto& [key, state] : daemons) {
+ f.open_object_section(ceph::to_string(key).c_str());
std::lock_guard l(state->lock);
for (auto ctr_inst_iter : state->perf_counters.instances) {
{
bool changed = false;
- lock.Lock();
+ lock.lock();
auto p = modules.find(module_name);
if (p != modules.end()) {
changed = p->second->set_health_checks(std::move(checks));
}
- lock.Unlock();
+ lock.unlock();
// immediately schedule a report to be sent to the monitors with the new
// health checks that have changed. This is done asynchronusly to avoid
}
int ActivePyModules::handle_command(
- std::string const &module_name,
+ const ModuleCommand& module_command,
+ const MgrSession& session,
const cmdmap_t &cmdmap,
const bufferlist &inbuf,
std::stringstream *ds,
std::stringstream *ss)
{
- lock.Lock();
- auto mod_iter = modules.find(module_name);
+ lock.lock();
+ auto mod_iter = modules.find(module_command.module_name);
if (mod_iter == modules.end()) {
- *ss << "Module '" << module_name << "' is not available";
+ *ss << "Module '" << module_command.module_name << "' is not available";
+ lock.unlock();
return -ENOENT;
}
- lock.Unlock();
- return mod_iter->second->handle_command(cmdmap, inbuf, ds, ss);
+ lock.unlock();
+ return mod_iter->second->handle_command(module_command, session, cmdmap,
+ inbuf, ds, ss);
}
void ActivePyModules::get_health_checks(health_check_map_t *checks)
{
std::lock_guard l(lock);
- for (auto& p : modules) {
- p.second->get_health_checks(checks);
+ for (auto& [name, module] : modules) {
+ dout(15) << "getting health checks for " << name << dendl;
+ module->get_health_checks(checks);
}
}
void ActivePyModules::config_notify()
{
std::lock_guard l(lock);
- for (auto& i : modules) {
- auto module = i.second.get();
+ for (auto& [name, module] : modules) {
// Send all python calls down a Finisher to avoid blocking
// C++ code, and avoid any potential lock cycles.
- finisher.queue(new FunctionContext([module](int r){
- module->config_notify();
- }));
+ dout(15) << "notify (config) " << name << dendl;
+ // workaround for https://bugs.llvm.org/show_bug.cgi?id=35984
+ finisher.queue(new LambdaContext([module=module](int r){
+ module->config_notify();
+ }));
}
}
dout(4) << " module " << module_name << " set URI '" << uri << "'" << dendl;
- modules[module_name]->set_uri(uri);
+ modules.at(module_name)->set_uri(uri);
}
-OSDPerfMetricQueryID ActivePyModules::add_osd_perf_query(
+MetricQueryID ActivePyModules::add_osd_perf_query(
const OSDPerfMetricQuery &query,
const std::optional<OSDPerfMetricLimit> &limit)
{
return server.add_osd_perf_query(query, limit);
}
-void ActivePyModules::remove_osd_perf_query(OSDPerfMetricQueryID query_id)
+void ActivePyModules::remove_osd_perf_query(MetricQueryID query_id)
{
int r = server.remove_osd_perf_query(query_id);
if (r < 0) {
}
}
-PyObject *ActivePyModules::get_osd_perf_counters(OSDPerfMetricQueryID query_id)
+PyObject *ActivePyModules::get_osd_perf_counters(MetricQueryID query_id)
{
std::map<OSDPerfMetricKey, PerformanceCounters> counters;
{
std::lock_guard l(lock);
- if (channel == "audit") {
- audit_clog->do_log(prio, message);
- } else {
- clog->do_log(prio, message);
- }
+ auto cl = monc.get_log_client()->create_channel(channel);
+ map<string,string> log_to_monitors;
+ map<string,string> log_to_syslog;
+ map<string,string> log_channel;
+ map<string,string> log_prio;
+ map<string,string> log_to_graylog;
+ map<string,string> log_to_graylog_host;
+ map<string,string> log_to_graylog_port;
+ uuid_d fsid;
+ string host;
+ if (parse_log_client_options(g_ceph_context, log_to_monitors, log_to_syslog,
+ log_channel, log_prio, log_to_graylog,
+ log_to_graylog_host, log_to_graylog_port,
+ fsid, host) == 0)
+ cl->update_config(log_to_monitors, log_to_syslog,
+ log_channel, log_prio, log_to_graylog,
+ log_to_graylog_host, log_to_graylog_port,
+ fsid, host);
+ cl->do_log(prio, message);
+}
+
+void ActivePyModules::register_client(std::string_view name, std::string addrs)
+{
+ std::lock_guard l(lock);
+
+ entity_addrvec_t addrv;
+ addrv.parse(addrs.data());
+
+ dout(7) << "registering msgr client handle " << addrv << dendl;
+ py_module_registry.register_client(name, std::move(addrv));
+}
+
+void ActivePyModules::unregister_client(std::string_view name, std::string addrs)
+{
+ std::lock_guard l(lock);
+
+ entity_addrvec_t addrv;
+ addrv.parse(addrs.data());
+
+ dout(7) << "unregistering msgr client handle " << addrv << dendl;
+ py_module_registry.unregister_client(name, addrv);
}
projects / ceph.git / blobdiff