Two defects in the same mechanism, both silent. 1. HibernateTxManager#commit fired synchronizations *after* its finally block, where cleanupIfIdle() had already called ResourceRegistry.cleanup() and removed the ThreadLocal list holding them. fireSynchronizations() then read a freshly initialized empty list and did nothing. No afterCommit callback had ever run on the Hibernate path: no exception, no log, just silence. rollback() twenty lines below had the order right, which is what makes this an ordering slip rather than a design choice. Found in production, from the far end: an admin write landed in Postgres while the in-memory cache it was registered to refresh never heard about it, so the change only took effect when the process restarted and re-read the database at boot. 2. Synchronizations were a single flat per-thread list, fired from index 0 by whichever transaction completed first. A REQUIRES_NEW inner transaction therefore fired the *suspended* outer transaction's callbacks too — early, with the inner transaction's outcome, for a transaction that might still roll back. Each new transaction now records how many synchronizations were already registered when it began, and fires only its own tail. Both managers get the fix and the same callback ordering: unbind the session or connection first, so a callback that opens its own transaction (a cache reload, an outbox drain) gets a fresh one instead of joining the transaction that just committed, then fire, then clean up. Also fixes JdbcTxStatus rejecting a null connection, which turned the two propagations that deliberately produce a connectionless status — SUPPORTS with no active transaction, and NOT_SUPPORTED — into an NPE inside begin(). The Hibernate manager always allowed it, and resource() already reports the real mistake with a message that names it. Tests: 16 new across the two managers, kept deliberately parallel since the two are interchangeable behind TxManager — synchronization firing, ordering, per-transaction scoping, callbacks opening their own transaction, and the previously untested SUPPORTS/NOT_SUPPORTED/MANDATORY propagations. Nothing covered afterCommit before, which is how both defects shipped. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Flash
A high-performance HTTP/1.1 server library for Java 21, built around virtual threads and a zero-allocation FSM router.
Modules
| Module | Description |
|---|---|
flash |
Core server library — router, request parser, HTTP I/O transport |
flash-extensions/flash-ext-jackson |
Jackson JSON integration |
flash-extensions/flash-ext-openapi |
OpenAPI 3.0 spec + Swagger UI |
flash-extensions/flash-ext-oidc |
OIDC Authorization Code + PKCE flow |
flash-extensions/flash-ext-mcp |
MCP (Model Context Protocol) server — Streamable HTTP, optional OAuth2 via flash-ext-oidc |
flash-extensions/flash-ext-view-core |
Minimal shared SSR runtime primitives |
flash-extensions/flash-ext-view-jte |
Opinionated jte SSR extension |
flash-extensions/flash-ext-view-thymeleaf |
Opinionated Thymeleaf SSR extension |
flash-bench |
Demo harness (OIDC + OpenAPI + Jackson) |
Requirements
- Java 21+
- Maven 3.8+
Quick start
FlashApp.create(8080)
.get("/ping", (req, res) -> "pong")
.start();
With full configuration:
FlashApp.create(
FlashConfiguration.builder()
.port(8080)
.host("0.0.0.0")
.maxHeaderBufferSize(65536)
.build()
)
.get("/ping", (req, res) -> "pong")
.start();
Route registration
Lambda routes
FlashApp app = FlashApp.create(8080);
app.get("/hello", (req, res) -> "world");
app.post("/echo", (req, res) -> {
byte[] body = req.body().bytes();
return res.status(200).body(body);
});
app.get("/users/{id}", (req, res) -> {
String id = req.pathParam("id");
return "user:" + id;
});
Class-based handlers
Extend RequestHandler, annotate it, then scan its package. Dependencies are cached in
onInit() after Flash has resolved its complete boot-time service graph:
@GET("/api/users")
public class ListUsers extends RequestHandler {
private UserService users;
@Override protected void onInit() { users = require(UserService.class); }
@Override public Object handle(Request req, Response res) { return users.list(); }
}
app.scan("dev.example.api");
Middleware
Apply middleware at registration. Flash composes the final chain at boot:
Middleware authCheck = next -> (req, res) -> {
if (req.header("Authorization") == null)
return res.status(401).body("Unauthorized");
return next.handle(req, res);
};
app.get("/secure", (req, res) -> "secret data", authCheck);
Multiple middlewares are composed outermost-first (left-to-right in the call):
app.get("/admin", handler, logging, auth, rateLimit);
// execution order: logging → auth → rateLimit → handler
Classpath scan
Scans a package for classes that extend RequestHandler and carry @Route. Each is
instantiated via its public no-arg constructor:
app.scan("dev.example.handlers");
Namespace mounting
Mount a scoped sub-router under a prefix. All routes registered inside the scope get the prefix prepended automatically. The scope inherits the parent's extension context (annotation processors, services):
app.mount("/api", scope -> {
scope.get("/health", (req, res) -> "ok"); // → GET /api/health
scope.scan("dev.example.api");
});
Extensions
Extensions have one declarative configure method. They declare services, processors and route
callbacks; Flash resolves the complete graph, materialises routes, compiles both routers, then
opens listeners. Extension install order never makes a service “not ready”.
FlashApp.create(8080)
.install(new JacksonExtension())
.install(new OpenApiExtension("/openapi", "My API", "1.0.0"))
.install(new OidcExtension(oidcConfig))
.scan("dev.example.handlers")
.start();
See extension-specific READMEs for full details:
flash-ext-jacksonflash-ext-openapiflash-ext-oidcflash-ext-mcpflash-ext-view-jteflash-ext-view-thymeleaf
Error handlers
app.onNotFound((req, res) -> res.status(404).body("Not found: " + req.path()));
app.onException((ex, req, res) -> {
if (ex instanceof IllegalArgumentException)
return res.status(400).body(ex.getMessage());
return res.status(500).body("Internal error");
});
FlashConfiguration
| Field | Default | Description |
|---|---|---|
port |
— | TCP port to bind |
host |
"0.0.0.0" |
Bind address |
tls |
null |
TLS for the default listener — see TLS |
listeners |
[] |
Multiple bind targets (port + host + optional TLS) on one app — see TLS |
maxHeaderBufferSize |
65536 |
Max size of the header buffer (bytes) |
TLS
HTTPS and WSS are a transport-layer concern only: once a listener is bound, the accepted
Socket is either plain or an SSLSocket indistinguishably from HttpServer's point of view
onward — the request parser, router, and WebSocket upgrade never branch on it. WSS is therefore
not a separate feature; it's a WebSocket upgrade running over whatever transport it was handed.
Quick start
FlashApp.create(FlashConfiguration.builder()
.port(443)
.tls(TlsConfig.keystore(Path.of("cert.p12"), "changeit"))
.build())
.get("/ping", (req, res) -> "pong") // HTTPS
.ws("/live", handler) // WSS, same route API
.start();
Multiple listeners
One app can bind any number of ports, each independently plain or TLS:
FlashApp.create(FlashConfiguration.builder()
.listener(new FlashConfiguration.Listener(80)) // plain
.listener(new FlashConfiguration.Listener(443, TlsConfig.keystore(cert, pass))) // TLS
.build());
A non-empty listeners list takes precedence over the top-level port/host/tls fields.
Each listener gets its own accept threads; the router, WS router, and virtual-thread executor
are shared by all of them — one app, N ports.
TlsConfig
| Factory | Use |
|---|---|
TlsConfig.keystore(Path, String) |
Builds the SSLContext from a PKCS12/JKS keystore (type guessed from the extension). Pins TLSv1.2/TLSv1.3 as enabled protocols; cipher suites are left at the JDK's own curated default. |
TlsConfig.ofContext(SSLContext) |
Escape hatch — the given SSLContext is used exactly as built. Flash never calls setSSLParameters on this path beyond what you explicitly request via clientAuth/applicationProtocols, so anything else you configured (custom KeyManager, ALPN, cipher suites) is authoritative. |
Chainable on either factory:
TlsConfig.keystore(cert, pass)
.clientAuth(ClientAuth.REQUIRE) // mTLS: NONE (default) | OPTIONAL | REQUIRE
.applicationProtocols("acme-tls/1", "http/1.1") // ALPN, in preference order
SNI falls out of keystore() for free: a keystore holding more than one certificate entry
is matched against the requested hostname by each certificate's SAN (falling back to CN) — no
per-hostname config. The first entry in the keystore is the default when SNI is absent or
matches nothing (same convention as nginx/HAProxy's default_server).
ALPN and custom certificate selection (e.g. TLS-ALPN-01 / RFC 8737 for on-demand ACME
issuance): ALPN is resolved while consuming ClientHello/producing ServerHello, which always
precedes Certificate production. A custom X509ExtendedKeyManager passed via ofContext
can therefore read engine.getHandshakeApplicationProtocol() (or
((SSLSocket) socket).getHandshakeApplicationProtocol()) inside
chooseEngineServerAlias/chooseServerAlias — the negotiated protocol is already resolved by
then, so the certificate decision can key off it.
mTLS with a private CA: clientAuth(...) only requests/requires a client certificate;
keystore() deliberately doesn't expose a way to configure which CAs are trusted for that
certificate (it uses the JDK default trust store). For a private CA, build the SSLContext
yourself with a TrustManagerFactory and use ofContext(...).
Reading TLS info from a request
app.get("/whoami", (req, res) -> {
if (!req.isSecure()) return "plain";
SSLSession session = req.sslSession(); // null iff !isSecure()
X509Certificate peer = (X509Certificate) session.getPeerCertificates()[0]; // mTLS only
return session.getCipherSuite() + " / " + session.getProtocol();
});
Request.isSecure() / Request.sslSession() cost nothing extra per request: the SSLSocket
reference is threaded through once per connection (same mechanism as remoteAddress()), and
sslSession() only calls SSLSocket#getSession() — a cached-field read once the handshake
that got the request this far has already completed, never a forced handshake.
WebSocketSession mirrors this exactly (isSecure(), sslSession()) by delegating to the
upgrading Request — no separate TLS state is tracked for WS.
Architecture
ServerSocket.accept()
→ RequestParser.parse() # zero-alloc header parsing, buffer reuse across keep-alive
→ GlobalRouter.route() # two-tier: mounted sub-routers (longest prefix) then FastPathRouterImpl
→ RequestHandler.handle() # user handler; return value sets body
→ Request.drain() # consume unread body for keep-alive
→ HttpServer writes response # status line, headers, then fixed or chunked body
→ loop or close socket # based on Connection header
- Virtual threads — each accepted socket runs on a virtual thread (
Executors.newVirtualThreadPerTaskExecutor()). Java 21 required. - Zero-allocation router —
FastPathRouterImplusesfpr-core, a byte-level FSM that matches onMETHOD + pathbytes with no per-request allocation. - Keep-alive —
RequestParserreuses its header buffer across requests on the same connection. - Chunked transfer — both chunked request bodies (decoded via
ChunkedInputStream) and chunked response bodies are supported. - TLS is transport-only — see TLS. Listeners bind either a plain
ServerSocketor anSSLServerSocket; nothing downstream ofaccept()branches on which.
Build & test
# Build all modules (skip tests)
mvn clean package -DskipTests
# Run all tests
mvn test
# Run a single test class
mvn test -pl flash -Dtest=RequestParserTest
# Run the benchmark demo server
java -jar flash-bench/target/flash-bench-1.0-SNAPSHOT.jar