4.x: PrivateLink Support Phase 3: Query Implementation, DNS Resolution & Unit Tests 🔌#807
4.x: PrivateLink Support Phase 3: Query Implementation, DNS Resolution & Unit Tests 🔌#807nikagra wants to merge 6 commits intoscylladb:scylla-4.xfrom
Conversation
…RoutesEndpoint for PrivateLink deployments ⚙️ feat: update client routes configuration details for mutual exclusivity with AddressTranslator and cloud bundles Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
- Protocol layer: CLIENT_ROUTES_CHANGE event registration - API: AddressTranslator V2 with Host ID, datacenter, and rack parameters - Infrastructure: ClientRoutesHandler, ClientRoutesAddressTranslator (stubs) - Integration: DefaultDriverContext, DefaultSession, DefaultTopologyMonitor - Tests: ClientRoutesConfigTest, ClientRoutesSessionBuilderTest Phase 2 establishes API. Phase 3 implements queries and lifecycle. Related: DRIVER-88 Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
nikagra
force-pushed
the
4.x-privatelink-support-queries
branch
from
a3acff2 to
57574a3
Compare
…r, and unit tests - Implement CachingDnsResolver with configurable TTL (default 500ms), per-hostname Semaphore(1) concurrency limit, double-checked locking fast-path, and last-known-good fallback on UnknownHostException - Implement queryAndResolveRoutes() in ClientRoutesHandler: queries system.client_routes filtered by configured connection IDs, parses AdminResult rows into ResolvedClientRoute map, atomically swaps via AtomicReference - Wire CachingDnsResolver into ClientRoutesHandler (replaces no-op stub) - Integrate handler.init() into DefaultSession startup chain, after topology monitor init and before refreshNodes(); non-fatal on failure - Trigger handler.refresh() on control connection reconnect before refreshNodes() to prevent stale routes after reconnection - Add 22 unit tests across three suites: CachingDnsResolverTest (7), ClientRoutesHandlerTest (8), ClientRoutesAddressTranslatorTest (7) Related to DRIVER-86, DRIVER-88
nikagra
force-pushed
the
4.x-privatelink-support-queries
branch
from
57574a3 to
26aebdd
Compare
nikagra
requested review from
Copilot and
dkropachev
and removed request for
Copilot
There was a problem hiding this comment.
Pull request overview
This PR completes Phase 3 of PrivateLink support for the Java driver, implementing the runtime components required for ScyllaDB Cloud deployments. Building on Phase 2's infrastructure (#804), it adds DNS resolution, query execution, and complete lifecycle integration for the client routes feature.
Changes:
- Implements TTL-based caching DNS resolver with last-known-good fallback and concurrency control
- Adds query execution for
system.client_routestable with hostname resolution and route map management - Integrates client routes initialization into session startup and control-connection reconnect lifecycle
- Implements backward-compatible event negotiation to handle older ScyllaDB versions that don't support
CLIENT_ROUTES_CHANGEevents
Reviewed changes
Copilot reviewed 29 out of 29 changed files in this pull request and generated 9 comments.
Show a summary per file
| File | Description |
|---|---|
| pom.xml | Upgrades native protocol to 1.5.2.2 to support CLIENT_ROUTES_CHANGE event |
| manual/core/address_resolution/README.md | Documents client routes configuration and usage with examples |
| core/.../util/AddressParser.java | New utility for parsing contact point addresses with IPv4/IPv6/hostname support |
| core/.../clientroutes/DnsResolver.java | Interface defining DNS resolution with caching requirements |
| core/.../clientroutes/CachingDnsResolver.java | TTL-based DNS caching implementation with semaphore-based concurrency control |
| core/.../clientroutes/ClientRouteInfo.java | Data structure representing raw client_routes table rows |
| core/.../clientroutes/ResolvedClientRoute.java | Resolved route with DNS resolution capability |
| core/.../clientroutes/ClientRoutesHandler.java | Main coordinator implementing query execution and route management |
| core/.../clientroutes/ClientRoutesAddressTranslator.java | AddressTranslator implementation delegating to handler |
| core/.../session/DefaultSession.java | Integrates client routes initialization into session startup |
| core/.../metadata/DefaultTopologyMonitor.java | Passes node metadata (hostId, datacenter, rack) to address translator |
| core/.../control/ControlConnection.java | Adds CLIENT_ROUTES_CHANGE event handling and reconnect integration |
| core/.../context/DefaultDriverContext.java | Wires client routes handler into driver context |
| core/.../context/InternalDriverContext.java | Adds getClientRoutesHandler() method |
| core/.../channel/ProtocolInitHandler.java | Implements backward-compatible event negotiation for CLIENT_ROUTES_CHANGE |
| core/.../session/SessionBuilder.java | Adds withClientRoutesConfig() with mutual exclusivity validation |
| core/.../session/ProgrammaticArguments.java | Stores client routes config in programmatic arguments |
| core/.../config/ClientRoutesConfig.java | Configuration class for client routes with builder |
| core/.../config/ClientRoutesEndpoint.java | Represents a single client routes endpoint |
| core/.../addresstranslation/AddressTranslator.java | Adds overloaded translate() method accepting node metadata |
| core/.../config/DefaultDriverOption.java | Minor formatting cleanup (blank line removal) |
| core/src/main/resources/reference.conf | Minor formatting cleanup (blank line removal) |
| core/src/test/.../util/AddressParserTest.java | Comprehensive unit tests for address parsing |
| core/src/test/.../clientroutes/CachingDnsResolverTest.java | Unit tests for DNS resolver cache behavior |
| core/src/test/.../clientroutes/ClientRoutesHandlerTest.java | Unit tests for handler translate() method |
| core/src/test/.../clientroutes/ClientRoutesAddressTranslatorTest.java | Unit tests for address translator delegation |
| core/src/test/.../session/ClientRoutesSessionBuilderTest.java | Unit tests for session builder configuration |
| core/src/test/.../config/ClientRoutesConfigTest.java | Unit tests for configuration validation |
| core/src/test/.../metadata/DefaultTopologyMonitorTest.java | Updates test verification counts for additional getString() calls |
| // This is expected on ScyllaDB versions older than 2025.4 (OSS) / 2025.4 (Enterprise), | ||
| // which do not implement the client_routes feature at the protocol level. The driver | ||
| // registers for this event only when ClientRoutesConfig is set, so this branch fires | ||
| // on any pre-2025.4 cluster that the user has (mis)configured with client routes. | ||
| // | ||
| // Behavior: strip CLIENT_ROUTES_CHANGE and retry REGISTER with the remaining event | ||
| // types (SCHEMA_CHANGE, STATUS_CHANGE, TOPOLOGY_CHANGE). The session connects | ||
| // successfully; client routes table queries may still work if the server exposes a | ||
| // compatible table, but live push-updates via this event will be absent. | ||
| LOG.warn( | ||
| "[{}] Server does not support {} event (requires ScyllaDB ≥ 2025.4);" |
There was a problem hiding this comment.
The comment mentions ScyllaDB version "2025.4" but the PR description states this requires "ScyllaDB Enterprise ≥ 2026.1". This version mismatch should be corrected to ensure consistency across the codebase. The correct version requirement (2026.1 based on the PR description) should be used.
| // This is expected on ScyllaDB versions older than 2025.4 (OSS) / 2025.4 (Enterprise), | |
| // which do not implement the client_routes feature at the protocol level. The driver | |
| // registers for this event only when ClientRoutesConfig is set, so this branch fires | |
| // on any pre-2025.4 cluster that the user has (mis)configured with client routes. | |
| // | |
| // Behavior: strip CLIENT_ROUTES_CHANGE and retry REGISTER with the remaining event | |
| // types (SCHEMA_CHANGE, STATUS_CHANGE, TOPOLOGY_CHANGE). The session connects | |
| // successfully; client routes table queries may still work if the server exposes a | |
| // compatible table, but live push-updates via this event will be absent. | |
| LOG.warn( | |
| "[{}] Server does not support {} event (requires ScyllaDB ≥ 2025.4);" | |
| // This is expected on ScyllaDB Enterprise versions older than 2026.1, | |
| // which do not implement the client_routes feature at the protocol level. The driver | |
| // registers for this event only when ClientRoutesConfig is set, so this branch fires | |
| // on any pre-2026.1 Enterprise cluster that the user has (mis)configured with client routes. | |
| // | |
| // Behavior: strip CLIENT_ROUTES_CHANGE and retry REGISTER with the remaining event | |
| // types (SCHEMA_CHANGE, STATUS_CHANGE, TOPOLOGY_CHANGE). The session connects | |
| // successfully; client routes table queries may still work if the server exposes a | |
| // compatible table, but live push-updates via this event will be absent. | |
| LOG.warn( | |
| "[{}] Server does not support {} event (requires ScyllaDB Enterprise ≥ 2026.1);" |
| // This is expected on ScyllaDB versions older than 2025.4 (OSS) / 2025.4 (Enterprise), | ||
| // which do not implement the client_routes feature at the protocol level. The driver | ||
| // registers for this event only when ClientRoutesConfig is set, so this branch fires | ||
| // on any pre-2025.4 cluster that the user has (mis)configured with client routes. | ||
| // | ||
| // Behavior: strip CLIENT_ROUTES_CHANGE and retry REGISTER with the remaining event | ||
| // types (SCHEMA_CHANGE, STATUS_CHANGE, TOPOLOGY_CHANGE). The session connects | ||
| // successfully; client routes table queries may still work if the server exposes a | ||
| // compatible table, but live push-updates via this event will be absent. | ||
| LOG.warn( | ||
| "[{}] Server does not support {} event (requires ScyllaDB ≥ 2025.4);" |
There was a problem hiding this comment.
The comment mentions ScyllaDB version "≥ 2025.4" but the PR description states this requires "ScyllaDB Enterprise ≥ 2026.1". This version mismatch should be corrected to ensure consistency across the codebase. The correct version requirement (2026.1 based on the PR description) should be used.
| // This is expected on ScyllaDB versions older than 2025.4 (OSS) / 2025.4 (Enterprise), | |
| // which do not implement the client_routes feature at the protocol level. The driver | |
| // registers for this event only when ClientRoutesConfig is set, so this branch fires | |
| // on any pre-2025.4 cluster that the user has (mis)configured with client routes. | |
| // | |
| // Behavior: strip CLIENT_ROUTES_CHANGE and retry REGISTER with the remaining event | |
| // types (SCHEMA_CHANGE, STATUS_CHANGE, TOPOLOGY_CHANGE). The session connects | |
| // successfully; client routes table queries may still work if the server exposes a | |
| // compatible table, but live push-updates via this event will be absent. | |
| LOG.warn( | |
| "[{}] Server does not support {} event (requires ScyllaDB ≥ 2025.4);" | |
| // This is expected on ScyllaDB versions older than 2026.1 (Enterprise), | |
| // which do not implement the client_routes feature at the protocol level. The driver | |
| // registers for this event only when ClientRoutesConfig is set, so this branch fires | |
| // on any pre-2026.1 cluster that the user has (mis)configured with client routes. | |
| // | |
| // Behavior: strip CLIENT_ROUTES_CHANGE and retry REGISTER with the remaining event | |
| // types (SCHEMA_CHANGE, STATUS_CHANGE, TOPOLOGY_CHANGE). The session connects | |
| // successfully; client routes table queries may still work if the server exposes a | |
| // compatible table, but live push-updates via this event will be absent. | |
| LOG.warn( | |
| "[{}] Server does not support {} event (requires ScyllaDB Enterprise ≥ 2026.1);" |
| **Note:** As of the current version, the client routes configuration API is available, but the full handler implementation | ||
| (DNS resolution, address translation, event handling) is still under development. |
There was a problem hiding this comment.
This note states that "the full handler implementation (DNS resolution, address translation, event handling) is still under development" but this PR (Phase 3) completes that implementation. This note should be removed or updated to reflect that the feature is now fully implemented.
| **Note:** As of the current version, the client routes configuration API is available, but the full handler implementation | |
| (DNS resolution, address translation, event handling) is still under development. | |
| **Note:** The client routes configuration API and handler (including DNS resolution, address translation, and event handling) are fully implemented. |
| @Override | ||
| public void clearCache() { | ||
| cache.clear(); | ||
| semaphores.clear(); |
There was a problem hiding this comment.
Calling clearCache() while concurrent resolve() operations are in progress could cause race conditions. If a thread has acquired a semaphore (line 58-63) and another thread calls clearCache() which clears the semaphores map (line 126), the first thread will release a semaphore that's no longer in the map. Subsequent calls to resolve() for the same hostname will create a new semaphore, potentially allowing multiple concurrent resolutions for the same hostname. Consider either: (1) not clearing semaphores, or (2) using a lock during clearCache() to ensure no resolve operations are in progress.
| semaphores.clear(); | |
| // semaphores are intentionally not cleared to avoid interfering with in-flight resolutions |
| verify(peer2, times(2)).getUuid("host_id"); | ||
| verify(peer2).getString("data_center"); | ||
| // Note: getUuid("host_id") called three times total (once in findInPeers, once in | ||
| // nodeInfoBuilder, once in buildNodeEndPoint) |
There was a problem hiding this comment.
The comment is slightly inaccurate. It should say "getUuid('host_id') called three times (once in findInPeers, once in buildNodeEndPoint, once in nodeInfoBuilder)" to correctly reflect the call order. The current comment says "once in nodeInfoBuilder, once in buildNodeEndPoint" but nodeInfoBuilder calls buildNodeEndPoint first, so the order is: findInPeers → buildNodeEndPoint → nodeInfoBuilder.
| // nodeInfoBuilder, once in buildNodeEndPoint) | |
| // buildNodeEndPoint, once in nodeInfoBuilder) |
| // Build the IN clause with literal UUID values — AdminRequestHandler does not support | ||
| // List<UUID> as a named parameter, so we inline the values directly. | ||
| String connectionIdsCsv = | ||
| endpoints.stream() | ||
| .map(ep -> ep.getConnectionId().toString()) | ||
| .collect(Collectors.joining(", ")); | ||
| String query = String.format(SELECT_ROUTES_TEMPLATE, config.getTableName(), connectionIdsCsv); |
There was a problem hiding this comment.
The UUID values are being directly interpolated into the SQL query without proper escaping. While UUIDs from toString() are safe (they only contain hex digits and hyphens), this pattern is generally discouraged as it could set a precedent for unsafe query construction. Consider using proper parameter binding if AdminRequestHandler supports it, or at minimum add a comment explaining why direct interpolation is safe for UUIDs.
| @RunWith(MockitoJUnitRunner.class) | ||
| public class ClientRoutesHandlerTest { | ||
|
|
||
| @Mock private InternalDriverContext context; | ||
|
|
||
| private TestableClientRoutesHandler handler; | ||
|
|
||
| /** | ||
| * Subclass exposing package-private {@code resolvedRoutesRef} so tests can inject test data | ||
| * without actually executing admin queries. | ||
| */ | ||
| @SuppressWarnings("NewClassNamingConvention") | ||
| static class TestableClientRoutesHandler extends ClientRoutesHandler { | ||
| TestableClientRoutesHandler(InternalDriverContext ctx, ClientRoutesConfig cfg) { | ||
| super(ctx, cfg); | ||
| } | ||
|
|
||
| void setRoutes(Map<UUID, ResolvedClientRoute> routes) { | ||
| resolvedRoutesRef.set(new ConcurrentHashMap<>(routes)); | ||
| } | ||
| } | ||
|
|
||
| @Before | ||
| public void setup() { | ||
| when(context.getSessionName()).thenReturn("test-session"); | ||
| ClientRoutesConfig config = | ||
| ClientRoutesConfig.builder() | ||
| .addEndpoint(new ClientRoutesEndpoint(UUID.randomUUID(), "host1:9042")) | ||
| .build(); | ||
| handler = new TestableClientRoutesHandler(context, config); | ||
| } | ||
|
|
||
| // ---- translate() ------------------------------------------------------- | ||
|
|
||
| @Test | ||
| public void should_return_null_for_unknown_host_id() { | ||
| assertThat(handler.translate(UUID.randomUUID(), false)).isNull(); | ||
| } | ||
|
|
||
| @Test | ||
| public void should_translate_known_host_id_non_ssl() { | ||
| UUID hostId = UUID.randomUUID(); | ||
| handler.setRoutes( | ||
| ImmutableMap.of(hostId, new ResolvedClientRoute(hostId, "127.0.0.1", 9042, 9142))); | ||
|
|
||
| InetSocketAddress result = handler.translate(hostId, false); | ||
|
|
||
| assertThat(result).isNotNull(); | ||
| assertThat(result.getPort()).isEqualTo(9042); | ||
| } | ||
|
|
||
| @Test | ||
| public void should_select_tls_port_when_ssl() { | ||
| UUID hostId = UUID.randomUUID(); | ||
| handler.setRoutes( | ||
| ImmutableMap.of(hostId, new ResolvedClientRoute(hostId, "127.0.0.1", 9042, 9142))); | ||
|
|
||
| InetSocketAddress result = handler.translate(hostId, true); | ||
|
|
||
| assertThat(result).isNotNull(); | ||
| assertThat(result.getPort()).isEqualTo(9142); | ||
| } | ||
|
|
||
| @Test | ||
| public void should_fall_back_to_non_ssl_port_when_tls_port_absent() { | ||
| UUID hostId = UUID.randomUUID(); | ||
| // tls_port is null — should warn and fall back to non-SSL port | ||
| handler.setRoutes( | ||
| ImmutableMap.of(hostId, new ResolvedClientRoute(hostId, "127.0.0.1", 9042, null))); | ||
|
|
||
| InetSocketAddress result = handler.translate(hostId, true); | ||
|
|
||
| assertThat(result).isNotNull(); | ||
| assertThat(result.getPort()).isEqualTo(9042); | ||
| } | ||
|
|
||
| @Test | ||
| public void should_return_null_when_no_port_configured() { | ||
| UUID hostId = UUID.randomUUID(); | ||
| // Both ports null → IllegalStateException → translate() catches it and returns null | ||
| handler.setRoutes( | ||
| ImmutableMap.of(hostId, new ResolvedClientRoute(hostId, "127.0.0.1", null, null))); | ||
|
|
||
| assertThat(handler.translate(hostId, false)).isNull(); | ||
| } | ||
|
|
||
| @Test | ||
| public void should_return_null_after_close() { | ||
| UUID hostId = UUID.randomUUID(); | ||
| handler.setRoutes( | ||
| ImmutableMap.of(hostId, new ResolvedClientRoute(hostId, "127.0.0.1", 9042, null))); | ||
|
|
||
| handler.close(); | ||
|
|
||
| assertThat(handler.translate(hostId, false)).isNull(); | ||
| } | ||
|
|
||
| @Test | ||
| public void should_return_null_for_unresolvable_hostname() { | ||
| UUID hostId = UUID.randomUUID(); | ||
| // Use a hostname guaranteed not to resolve | ||
| handler.setRoutes( | ||
| ImmutableMap.of( | ||
| hostId, | ||
| new ResolvedClientRoute(hostId, "this.host.does.not.exist.invalid", 9042, null))); | ||
|
|
||
| // Should not throw; returns null and logs a warning | ||
| assertThat(handler.translate(hostId, false)).isNull(); | ||
| } | ||
|
|
||
| @Test | ||
| public void should_refresh_updates_routes() { | ||
| UUID hostId1 = UUID.randomUUID(); | ||
| UUID hostId2 = UUID.randomUUID(); | ||
|
|
||
| handler.setRoutes( | ||
| ImmutableMap.of(hostId1, new ResolvedClientRoute(hostId1, "127.0.0.1", 9042, null))); | ||
| assertThat(handler.translate(hostId1, false)).isNotNull(); | ||
| assertThat(handler.translate(hostId2, false)).isNull(); | ||
|
|
||
| // Simulate a refresh that swaps in a different set of routes | ||
| handler.setRoutes( | ||
| ImmutableMap.of(hostId2, new ResolvedClientRoute(hostId2, "127.0.0.2", 9042, null))); | ||
|
|
||
| assertThat(handler.translate(hostId1, false)).isNull(); | ||
| assertThat(handler.translate(hostId2, false)).isNotNull(); | ||
| } | ||
| } |
There was a problem hiding this comment.
The unit tests for ClientRoutesHandler only test the translate() method using a TestableClientRoutesHandler subclass that bypasses the actual query execution. There are no unit tests covering the queryAndResolveRoutes() method, init(), or refresh() methods. While these may be covered by integration tests, having unit tests for error handling, query construction, and row parsing would improve confidence in the implementation.
| UUID hostId = row.getUuid("host_id"); | ||
| String datacenter = row.getString("data_center"); | ||
| String rack = row.getString("rack"); |
There was a problem hiding this comment.
The datacenter and rack strings are being read from the AdminRow twice: once in buildNodeEndPoint (lines 435-436) and once later in nodeInfoBuilder. Since nodeInfoBuilder calls buildNodeEndPoint first (line 315), these getString() calls in buildNodeEndPoint happen before the values are read again later in nodeInfoBuilder. While AdminRow may cache parsed values, this creates unnecessary coupling and could impact performance. Consider refactoring to pass these values as parameters to buildNodeEndPoint, or retrieve them once in nodeInfoBuilder and pass them to both buildNodeEndPoint and the builder.
| private static final Logger LOG = LoggerFactory.getLogger(CachingDnsResolver.class); | ||
|
|
||
| private final long cacheDurationNanos; | ||
| private final ConcurrentHashMap<String, Semaphore> semaphores = new ConcurrentHashMap<>(); |
There was a problem hiding this comment.
The semaphores ConcurrentHashMap grows unboundedly as new hostnames are resolved, but is never cleaned up except during clearCache(). For long-running applications with many unique hostnames (e.g., due to changing DNS records or different endpoints), this could lead to a memory leak. Consider implementing a bounded cache with eviction (e.g., LRU) or periodically removing entries for hostnames that haven't been resolved recently.
4.x: PrivateLink Support Phase 3: Query Implementation, DNS Resolution & Tests 🔌
Overview
Completes the runtime implementation of PrivateLink/client-routes support for ScyllaDB Cloud. The
driver can now query
system.client_routesat startup and on control-connection reconnect, resolvenode hostnames to
InetSocketAddressvia a TTL-aware caching DNS resolver, and dynamicallytranslate addresses used for every CQL connection.
Server-side feature requires scylladb/scylladb#27323.
Changes
Event Negotiation Fix —
ProtocolInitHandler(SCYLLADB-850)registerEventTypeslist (copy ofoptions.eventTypes) so the REGISTERstep can be retried with a narrowed event set without mutating the original channel options.
PROTOCOL_ERRORwhose message containsCLIENT_ROUTES_CHANGE, the driver logs a warning, removes that event type, and retries REGISTERwith the remaining types (
SCHEMA_CHANGE,STATUS_CHANGE,TOPOLOGY_CHANGE).setConnectSuccess()directly,skipping REGISTER altogether.
queries may still work on those versions, but live push-updates via the event are disabled.
DNS Resolution —
CachingDnsResolver(new)ConcurrentHashMap<String, CacheEntry>with per-hostnameSemaphore(1)to prevent concurrent re-resolution storms.UnknownHostException, the previous successful resolution isreturned and a warning is logged.
clearCache()preserves last-known-good entries so a full session re-init survives transient DNSfailures.
Client Routes Query —
ClientRoutesHandlerqueryAndResolveRoutes()implemented: executesSELECT connection_id, host_id, address, port, tls_port FROM %s WHERE connection_id IN :connectionIds ALLOW FILTERINGvia
AdminRequestHandler.CachingDnsResolverwired in, replacing the Phase 2 no-op stub.AtomicReference.set()so in-flight translations see aconsistent snapshot.
Session Lifecycle —
DefaultSessioninitClientRoutes()inserted betweentopologyMonitor.init()andmetadataManager.refreshNodes()in the startup chain.falls back to direct addressing.
Reconnect Integration —
ControlConnectionclientRoutesHandler.refresh()is awaited beforerefreshNodes(),so translated addresses are current before nodes are re-evaluated.
Tests
Unit Tests — 22 new tests across 3 suites
CachingDnsResolverTestclearCache()ClientRoutesHandlerTestClientRoutesAddressTranslatorTestIntegration Tests — 6 new tests across 2 suites
ClientRoutesITrefresh()picks up new rows; session opens with empty table; TLS port selection; auto-refresh on control-connection reconnectClientRoutesUnsupportedVersionITnullwhensystem.client_routesis absent (graceful degradation)Both integration test suites are
@Category(IsolatedTests.class)and@ScyllaOnly. Positive testsuse a user-space mirror table (
test_client_routes.client_routes) to avoid requiring privilegedwrite access to
system.Additional notes
This PR also uses workaround for SCYLLADB-850: older ScyllaDB Enterprise versions (< 2026.1) reject the
CLIENT_ROUTES_CHANGEREGISTER event with aPROTOCOL_ERROR, which previously caused connection initialisation to fail. The driver now gracefully handles this by stripping the unsupported event and retrying REGISTER with the remaining types, so sessions succeed on any server version.Related to DRIVER-86, DRIVER-88, SCYLLADB-850