Files
logos-protocol/cpp/logos_api_client.cpp
T
Dario LipicarandClaude Opus 5 0183e8c26a fix: make the caller's timeout bound the token handshake, not just the call (#55)
* test(token): pin that the caller's budget bounds the handshake, not just the call

Written first and on its own commit so the red is on the record: against this
parent it fails at ~20s, because the capability handshake ignores the caller's
budget entirely.

LogosAPIClient::invokeRemoteMethod takes a Timeout, but on an un-tokened target
the handshake runs FIRST and LogosAPIConsumer::requestModule hardcodes 20000
twice -- once for the capability_module acquire, once for the requestModule call
on it. A caller asking for 1500ms could therefore block on the order of 40s
before the part it had actually bounded began. logos-view-module-runtime's
callModule advertises a 1500ms bound on precisely this path.

capability_module is deliberately NOT published, so the acquire runs its budget
out rather than succeeding. Every other test in this file publishes it, which is
how a hardcoded 20s survived alongside them: none of them ever entered the wait.

The assertion is two-sided on purpose. An upper bound alone would pass if
something made the acquire return instantly -- leaving the hardcoded 20s in
place and the test green for the wrong reason, which is the exact shape of two
earlier tests in this change set that passed in both directions. So:

  >= budget-200ms  proves the timeout path actually ran;
  <  4x budget     proves it was the CALLER's budget and not the 20s default.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix: make the caller's timeout bound the token handshake, not just the call

requestModule gains a timeoutMs parameter (defaulted to today's 20000, so every
existing caller is source-compatible and unchanged), and LogosAPIClient threads
its caller's Timeout through mintAndCacheToken into it.

The budget bounds the WHOLE handshake -- the capability_module acquire plus the
requestModule call on it share one deadline, rather than each getting a fresh
copy. Halving it would be arbitrary; giving each the full amount would make the
worst case twice what the caller asked for. What is left after the acquire is
never allowed to reach 0, because some transports read 0 as "no timeout" and an
exhausted budget must not silently become an unbounded wait.

Also corrects a comment that argued the handshake-refusal fallthrough was safe
because "capability_module passes 3000 ms". It does not: capability_module
reaches informModuleToken_module through its FOUR-argument overload
(capability_module_plugin.cpp:112), so timeoutMs takes the header's 20 s
default. The bound is real, it is just not short -- and the code should say the
true thing about why it is safe.

Not covered here: the ASYNC first-call path still acquires capability_module
through invokeRemoteMethodAsync without threading a budget. It does not block
the caller, so it is a different defect with a different fix.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 14:12:23 -03:00

532 lines
25 KiB
C++

#include "logos_api_client.h"
#include "logos_api_consumer.h"
#include "logos_object.h"
#include "logos_types.h"
#include "logos_json_convert.h"
#include "logos_thread_marshal.h"
#include "logos_rpc_status.h"
#include "token_manager.h"
#include <QJsonDocument>
#include <QJsonObject>
#include <QJsonArray>
#include <QJsonValue>
#include <QMetaObject>
#include <QMetaType>
#include <QPointer>
#include <string>
using logos::qvariantToNlohmann;
using logos::nlohmannArgsToQVariantList;
LogosAPIClient::LogosAPIClient(const QString& module_to_talk_to,
const QString& origin_module,
TokenManager* token_manager,
const LogosTransportConfig& target_transport,
const LogosTransportConfig& capability_transport,
QObject *parent)
: QObject(parent)
, m_consumer(new LogosAPIConsumer(module_to_talk_to, origin_module,
token_manager, target_transport, this))
, m_token_manager(token_manager)
, m_origin_module(origin_module)
// Pre-build the capability_module consumer once. We skip it for
// the capability_module client itself — the auto-`requestModule`
// path is gated by `objectName != "capability_module"` so we'd
// never use it, and constructing one would be a redundant
// self-connection. Init-list order matches the declaration order
// in the header — `m_capability_consumer` is appended at the end
// for ABI stability (see header comment).
, m_capability_consumer(module_to_talk_to == QStringLiteral("capability_module")
? nullptr
: new LogosAPIConsumer(QStringLiteral("capability_module"),
origin_module, token_manager,
capability_transport, this))
{
}
LogosAPIClient::LogosAPIClient(const QString& module_to_talk_to,
const QString& origin_module,
TokenManager* token_manager,
QObject *parent)
: LogosAPIClient(module_to_talk_to, origin_module, token_manager,
LogosTransportConfigGlobal::getDefault(),
LogosTransportConfigGlobal::getDefault(), parent)
{
}
LogosAPIClient::~LogosAPIClient()
{
}
LogosObject* LogosAPIClient::requestObject(const QString& objectName, Timeout timeout)
{
// Marshal to the owner thread: the replica is acquired and lives there.
return logos::runOnOwnerThread(this, [&]() -> LogosObject* {
return m_consumer->requestObject(objectName, timeout);
});
}
bool LogosAPIClient::isConnected() const
{
return m_consumer->isConnected();
}
QString LogosAPIClient::registryUrl() const
{
return m_consumer->registryUrl();
}
bool LogosAPIClient::reconnect()
{
return m_consumer->reconnect();
}
QVariant LogosAPIClient::invokeRemoteMethod(const QString& objectName, const QString& methodName,
const QVariantList& args, Timeout timeout)
{
return invokeRemoteMethod(objectName, methodName, args, timeout, nullptr);
}
QVariant LogosAPIClient::invokeRemoteMethod(const QString& objectName, const QString& methodName,
const QVariantList& args, Timeout timeout, logos::CallError* err)
{
if (err) err->clear();
// Marshal the whole operation (capability/token fetch + the call) onto the
// owner thread so a worker thread (e.g. an HTTP handler) can call other
// modules. Same-thread callers run directly. See logos_thread_marshal.h.
return logos::runOnOwnerThread(this, [&]() -> QVariant {
qDebug() << "LogosAPIClient: invoking remote method" << objectName << methodName << "args_count:" << args.size();
const bool eligible = objectName != QStringLiteral("capability_module") && m_capability_consumer;
QString token = getToken(objectName);
if (token.isEmpty() && eligible)
token = mintAndCacheToken(objectName, timeout); // first exchange (cached for later calls)
QVariant result = m_consumer->invokeRemoteMethod(token, objectName, methodName, args, timeout, err);
// Re-exchange on rejection, once. The provider rejected our (stale) token —
// drop it, mint a fresh one via capability_module, and retry the call. Gated
// on the explicit provider sentinel (never a plain empty result), so it can't
// loop, can't misfire on a legitimately-empty return, and never fires against
// an old provider (which returns a bare QVariant() we don't match). This also
// lazily recovers the common provider-reload case. See logos_rpc_status.h.
if (eligible && logos::isUnauthorizedSentinel(result)) {
qWarning() << "LogosAPIClient: token for" << objectName
<< "rejected by provider; re-exchanging and retrying once";
m_token_manager->removeToken(objectName);
const QString fresh = mintAndCacheToken(objectName, timeout);
if (!fresh.isEmpty())
result = m_consumer->invokeRemoteMethod(fresh, objectName, methodName, args, timeout, err);
}
// Never surface the sentinel to the typed wrapper. If we still hold it the
// retry failed (capability down / provider truly gone): collapse to today's
// empty result, and for NEW callers set a distinguishable CallError.
if (logos::isUnauthorizedSentinel(result)) {
if (err) {
err->code = "unauthorized";
err->message = "call to '" + objectName.toStdString()
+ "' rejected: token not recognized (re-exchange failed)";
err->origin = objectName.toStdString();
}
return QVariant();
}
return result;
});
}
QString LogosAPIClient::mintAndCacheToken(const QString& objectName, Timeout timeout)
{
qDebug() << "LogosAPIClient: calling requestModule for" << objectName;
const QString capabilityToken = getToken(QStringLiteral("capability_module"));
const QString token = QString::fromStdString(
m_capability_consumer->requestModule(capabilityToken.toStdString(),
m_origin_module.toStdString(),
objectName.toStdString(),
timeout.ms));
qDebug() << "LogosAPIClient: requestModule result for" << objectName << ":" << token;
// Cache the minted token so subsequent calls skip the handshake — closes the
// token-rotation race where overlapping requestModule calls mint fresh tokens
// that overwrite each other at the target (e.g. QtRO's sync wait reentering
// via a nested event loop).
if (!token.isEmpty())
m_token_manager->saveToken(objectName, token);
return token;
}
QVariant LogosAPIClient::invokeRemoteMethod(const QString& objectName, const QString& methodName,
const QVariant& arg, Timeout timeout)
{
return invokeRemoteMethod(objectName, methodName, QVariantList() << arg, timeout);
}
QVariant LogosAPIClient::invokeRemoteMethod(const QString& objectName, const QString& methodName,
const QVariant& arg1, const QVariant& arg2, Timeout timeout)
{
return invokeRemoteMethod(objectName, methodName, QVariantList() << arg1 << arg2, timeout);
}
QVariant LogosAPIClient::invokeRemoteMethod(const QString& objectName, const QString& methodName,
const QVariant& arg1, const QVariant& arg2, const QVariant& arg3, Timeout timeout)
{
return invokeRemoteMethod(objectName, methodName, QVariantList() << arg1 << arg2 << arg3, timeout);
}
QVariant LogosAPIClient::invokeRemoteMethod(const QString& objectName, const QString& methodName,
const QVariant& arg1, const QVariant& arg2, const QVariant& arg3,
const QVariant& arg4, Timeout timeout)
{
return invokeRemoteMethod(objectName, methodName, QVariantList() << arg1 << arg2 << arg3 << arg4, timeout);
}
QVariant LogosAPIClient::invokeRemoteMethod(const QString& objectName, const QString& methodName,
const QVariant& arg1, const QVariant& arg2, const QVariant& arg3,
const QVariant& arg4, const QVariant& arg5, Timeout timeout)
{
return invokeRemoteMethod(objectName, methodName, QVariantList() << arg1 << arg2 << arg3 << arg4 << arg5, timeout);
}
void LogosAPIClient::invokeRemoteMethodAsync(const QString& objectName, const QString& methodName,
const QVariantList& args, AsyncResultCallback callback,
Timeout timeout)
{
// Delegate to the CallError-aware overload; legacy callers just drop the
// error field. Keeps the handshake-coalescing logic single-sourced.
invokeRemoteMethodAsync(objectName, methodName, args,
[cb = std::move(callback)](QVariant r, const logos::CallError&) mutable {
if (cb) cb(std::move(r));
},
timeout);
}
void LogosAPIClient::invokeRemoteMethodAsync(const QString& objectName, const QString& methodName,
const QVariantList& args, AsyncResultErrorCallback callback,
Timeout timeout)
{
// Public entry: grant one retry for the rejection-driven re-exchange.
invokeRemoteMethodAsyncImpl(objectName, methodName, args, std::move(callback), timeout, /*retriesLeft=*/1);
}
void LogosAPIClient::invokeRemoteMethodAsyncImpl(const QString& objectName, const QString& methodName,
const QVariantList& args, AsyncResultErrorCallback callback,
Timeout timeout, int retriesLeft)
{
if (!callback) return;
// The async path acquires a replica too, so it must also run on the owner
// thread. Unlike the sync path we post non-blocking (QueuedConnection): the
// worker caller returns immediately and the result callback fires on the
// owner thread when the reply arrives. Preserve retriesLeft across the hop.
if (QThread::currentThread() != this->thread()) {
QMetaObject::invokeMethod(this,
[this, objectName, methodName, args,
callback = std::move(callback), timeout, retriesLeft]() mutable {
invokeRemoteMethodAsyncImpl(objectName, methodName, args,
std::move(callback), timeout, retriesLeft);
},
Qt::QueuedConnection);
return;
}
const bool eligible = objectName != QStringLiteral("capability_module") && m_capability_consumer;
// Wrap the user callback so a provider rejection sentinel triggers one
// re-exchange + retry, and the sentinel is never surfaced to the caller.
// Mirrors the sync path's retry in logos_api_client.cpp's invokeRemoteMethod.
QPointer<LogosAPIClient> selfGuard(this);
AsyncResultErrorCallback onResult =
[this, selfGuard, objectName, methodName, args, timeout, retriesLeft, cb = std::move(callback)]
(QVariant result, const logos::CallError& err) mutable {
if (!selfGuard) return; // client destroyed mid-flight: drop
if (retriesLeft > 0 && objectName != QStringLiteral("capability_module")
&& m_capability_consumer && logos::isUnauthorizedSentinel(result)) {
qWarning() << "LogosAPIClient: token for" << objectName
<< "rejected by provider (async); re-exchanging and retrying once";
m_token_manager->removeToken(objectName);
// Token is empty now → the re-entry coalesces the retry through the
// same m_pendingHandshakes machinery, so a burst of concurrent
// rejections doesn't restorm capability_module with N handshakes.
invokeRemoteMethodAsyncImpl(objectName, methodName, args,
std::move(cb), timeout, retriesLeft - 1);
return;
}
if (logos::isUnauthorizedSentinel(result)) {
logos::CallError e;
e.code = "unauthorized";
e.message = "call to '" + objectName.toStdString()
+ "' rejected: token not recognized (re-exchange failed)";
e.origin = objectName.toStdString();
cb(QVariant(), e);
return;
}
cb(std::move(result), err);
};
QString token = getToken(objectName);
if (token.isEmpty() && eligible) {
// Async-chain: dispatch the requestModule call asynchronously, and only
// fire the real method's invokeRemoteMethodAsync from its callback. The
// previous version called `requestModule` synchronously here, which made
// the "async" entry point block its caller for the full round-trip.
//
// COALESCE concurrent first-calls behind ONE handshake. A driver that
// fans out N async calls to an un-tokened target before any completes
// would otherwise fire N separate requestModule handshakes; each mints a
// distinct token and informs the target, and the later inform OVERWRITES
// the earlier token there (the target stores one token per caller). The
// already-dispatched calls then carry a superseded token and the target
// rejects them as unauthorized. So only the first caller starts the
// handshake; the rest queue and all drain with the single minted token.
// (The sync path can't hit this — it blocks per call, so handshakes
// never overlap.) m_pendingHandshakes is touched only on the owner
// thread, reached above, so no lock is needed.
m_pendingHandshakes[objectName].push_back(
[this, objectName, methodName, args, timeout, cb = std::move(onResult)]
(const QString& tok) mutable {
m_consumer->invokeRemoteMethodAsync(tok, objectName, methodName, args,
std::move(cb), timeout);
});
if (m_pendingHandshakes[objectName].size() > 1)
return; // a handshake for this target is already in flight
const QString capabilityToken = getToken("capability_module");
const QString origin = m_origin_module;
// Lifetime: capture the client through a QPointer guard. If it (and its
// QObject-parented consumers + the pending queue) is destroyed while the
// requestModule round-trip is in flight, the guard goes null and we drop
// the queued continuations instead of dereferencing dangling memory.
QPointer<LogosAPIClient> self(this);
m_capability_consumer->invokeRemoteMethodAsync(
capabilityToken,
QStringLiteral("capability_module"),
QStringLiteral("requestModule"),
QVariantList() << origin << objectName,
[self, objectName](const QVariant& tokenResult) mutable {
if (!self) return; // client destroyed mid-flight
const QString tok = tokenResult.toString();
// Cache the minted token before draining so future calls skip the handshake — m_pendingHandshakes only coalesces the first burst, the cache stops a second burst from racing the same rotation.
if (!tok.isEmpty()) self->m_token_manager->saveToken(objectName, tok);
// Drain every continuation queued for this target with the one
// minted token — the target was informed of exactly this token.
// An empty tok (handshake failed) still flows through: the
// consumer call is then rejected and each callback fires with an
// invalid QVariant, so callers never hang.
auto it = self->m_pendingHandshakes.find(objectName);
if (it == self->m_pendingHandshakes.end()) return;
std::vector<std::function<void(const QString&)>> calls = std::move(it.value());
self->m_pendingHandshakes.erase(it);
for (auto& c : calls) c(tok);
},
timeout);
return;
}
m_consumer->invokeRemoteMethodAsync(token, objectName, methodName, args, std::move(onResult), timeout);
}
void LogosAPIClient::invokeRemoteMethodAsync(const QString& objectName, const QString& methodName,
const QVariant& arg, AsyncResultCallback callback,
Timeout timeout)
{
invokeRemoteMethodAsync(objectName, methodName, QVariantList() << arg, std::move(callback), timeout);
}
void LogosAPIClient::invokeRemoteMethodAsync(const QString& objectName, const QString& methodName,
const QVariant& arg1, const QVariant& arg2,
AsyncResultCallback callback, Timeout timeout)
{
invokeRemoteMethodAsync(objectName, methodName, QVariantList() << arg1 << arg2, std::move(callback), timeout);
}
void LogosAPIClient::invokeRemoteMethodAsync(const QString& objectName, const QString& methodName,
const QVariant& arg1, const QVariant& arg2, const QVariant& arg3,
AsyncResultCallback callback, Timeout timeout)
{
invokeRemoteMethodAsync(objectName, methodName, QVariantList() << arg1 << arg2 << arg3, std::move(callback), timeout);
}
void LogosAPIClient::invokeRemoteMethodAsync(const QString& objectName, const QString& methodName,
const QVariant& arg1, const QVariant& arg2, const QVariant& arg3,
const QVariant& arg4, AsyncResultCallback callback,
Timeout timeout)
{
invokeRemoteMethodAsync(objectName, methodName, QVariantList() << arg1 << arg2 << arg3 << arg4, std::move(callback), timeout);
}
void LogosAPIClient::invokeRemoteMethodAsync(const QString& objectName, const QString& methodName,
const QVariant& arg1, const QVariant& arg2, const QVariant& arg3,
const QVariant& arg4, const QVariant& arg5,
AsyncResultCallback callback, Timeout timeout)
{
invokeRemoteMethodAsync(objectName, methodName, QVariantList() << arg1 << arg2 << arg3 << arg4 << arg5, std::move(callback), timeout);
}
void LogosAPIClient::onEvent(LogosObject* originObject, const QString& eventName, std::function<void(const QString&, const QVariantList&)> callback)
{
// Marshal to the owner thread: event registration touches the replica.
logos::runOnOwnerThread(this, [&]() {
m_consumer->onEvent(originObject, eventName, std::move(callback));
});
}
quint64 LogosAPIClient::onEventWhenAvailable(const QString& objectName, const QString& eventName,
std::function<void(const QString&, const QVariantList&)> callback,
std::function<void(bool)> onArmed)
{
// Marshal to the owner thread for the same reason onEvent() does: the
// registry touches (and later arms against) a QtRO replica, which only
// works on the thread that created the node.
return logos::runOnOwnerThread(this, [&]() -> quint64 {
return m_consumer->onEventWhenAvailable(objectName, eventName,
std::move(callback), std::move(onArmed));
});
}
quint64 LogosAPIClient::whenObjectAvailable(const QString& objectName,
std::function<void(bool)> onReady)
{
// Same owner-thread marshalling as onEventWhenAvailable: the registry
// touches a QtRO node that only works on the thread that created it.
return logos::runOnOwnerThread(this, [&]() -> quint64 {
return m_consumer->whenObjectAvailable(objectName, std::move(onReady));
});
}
bool LogosAPIClient::cancelEventSubscription(quint64 subscriptionId)
{
return logos::runOnOwnerThread(this, [&]() -> bool {
return m_consumer->cancelEventSubscription(subscriptionId);
});
}
LogosSubscriptionState LogosAPIClient::eventSubscriptionState(quint64 subscriptionId) const
{
return logos::runOnOwnerThread(const_cast<LogosAPIClient*>(this),
[&]() -> LogosSubscriptionState {
return m_consumer->eventSubscriptionState(subscriptionId);
});
}
QStringList LogosAPIClient::pendingEventSubscriptions() const
{
return logos::runOnOwnerThread(const_cast<LogosAPIClient*>(this), [&]() -> QStringList {
return m_consumer->pendingSubscriptions();
});
}
void LogosAPIClient::onEventResponse(LogosObject* object, const QString& eventName, const QVariantList& data)
{
qDebug() << "[LogosObject] LogosAPIClient::onEventResponse" << eventName << "-> LogosObject::emitEvent";
if (eventName.isEmpty()) {
qWarning() << "LogosAPIClient: Event name cannot be empty";
return;
}
if (!object) {
qWarning() << "LogosAPIClient: Cannot emit event on null object";
return;
}
object->emitEvent(eventName, data);
}
void LogosAPIClient::onEventResponse(QObject* object, const QString& eventName, const QVariantList& data)
{
qDebug() << "[LogosObject] LogosAPIClient::onEventResponse (QObject* compat)" << eventName;
if (eventName.isEmpty()) {
qWarning() << "LogosAPIClient: Event name cannot be empty";
return;
}
if (!object) {
qWarning() << "LogosAPIClient: Cannot emit event on null QObject";
return;
}
QMetaObject::invokeMethod(object, "eventResponse",
Qt::DirectConnection,
Q_ARG(QString, eventName),
Q_ARG(QVariantList, data));
}
bool LogosAPIClient::informModuleToken(const QString& authToken, const QString& moduleName, const QString& token)
{
return m_consumer->informModuleToken(authToken, moduleName, token);
}
bool LogosAPIClient::informModuleToken(const std::string& authToken, const std::string& moduleName, const std::string& token)
{
return informModuleToken(QString::fromStdString(authToken),
QString::fromStdString(moduleName),
QString::fromStdString(token));
}
bool LogosAPIClient::informModuleToken_module(const QString& authToken, const QString& originModule, const QString& moduleName, const QString& token, int timeoutMs)
{
// Marshal to the owner thread, exactly as requestObject/invokeRemoteMethod do.
// This path now goes through acquireCachedObject, so it reads and mutates
// m_objectCache — declared single-threaded, and holding thread-affine QtRO
// handles. Before the handshake surface existed this method used an
// uncached requestObject + release(), so it touched no shared state; routing
// it onto the cache is what made the missing marshal reachable.
//
// (LogosAPIClient::informModuleToken — the 3-arg form above — has the same
// missing marshal, but it still uses an uncached handle and predates this
// change, so it is left alone rather than widened into this fix.)
return logos::runOnOwnerThread(this, [&]() -> bool {
return m_consumer->informModuleToken_module(authToken, originModule, moduleName, token, timeoutMs);
});
}
TokenManager* LogosAPIClient::getTokenManager() const
{
return m_token_manager;
}
QString LogosAPIClient::getToken(const QString& module_name)
{
qDebug() << "LogosAPIClient: getToken for module:" << module_name;
QString token = m_token_manager->getToken(module_name);
if (!token.isEmpty()) {
qDebug() << "LogosAPIClient: Found token for module:" << module_name;
return token;
}
qDebug() << "LogosAPIClient: No token found for module:" << module_name;
return "";
}
// ---------------------------------------------------------------------------
// nlohmann::json overloads
// ---------------------------------------------------------------------------
nlohmann::json LogosAPIClient::invokeRemoteMethod(const std::string& objectName,
const std::string& methodName,
const nlohmann::json& args,
Timeout timeout)
{
QVariantList qArgs = nlohmannArgsToQVariantList(args);
QVariant result = invokeRemoteMethod(
QString::fromStdString(objectName),
QString::fromStdString(methodName),
qArgs, timeout);
return qvariantToNlohmann(result);
}
void LogosAPIClient::onEvent(LogosObject* originObject, const std::string& eventName,
std::function<void(const std::string&, const nlohmann::json&)> callback)
{
onEvent(originObject, QString::fromStdString(eventName),
[cb = std::move(callback)](const QString& name, const QVariantList& data) {
nlohmann::json jData = nlohmann::json::array();
for (const QVariant& v : data)
jData.push_back(qvariantToNlohmann(v));
cb(name.toStdString(), jData);
});
}