Class: NewBridge::SessionClient
- Inherits:
-
Object
- Object
- NewBridge::SessionClient
- Defined in:
- lib/new_bridge/session_client.rb
Overview
NewBridge::SessionClient
Responsibilities (Ruby side, inside one Ruby process):
- Accept one connection from the R gatekeeper (Unix socket or TCP).
- Serialize and send framed MsgPack envelopes of type
REQ. - Wait for matching framed MsgPack envelopes of type
RET. - Handle R -> Ruby callback requests (
CALL) by invoking registered Ruby blocks in background threads and replying withRET.
Protocol basics
- Transport: Unix domain socket for local R (loopback), TCP when R connects to a non-local host (e.g. Docker). Override with ENV = unix|tcp. Same MsgPack framing.
- Envelopes are MsgPack maps handled by
NewBridge::Envelope. - Envelope fields used here:
call_id: UUID used to correlate one REQ with one RETtype:"REQ"/"RET"/"CALL"session_id: logical session; R uses it to isolate per-session env stateinstance_id: logical instance routing; used for multi-instance poolsparent_id: reserved for nested-call correlation (Phase 4+)
Threading model
- One reader thread (
reader_loop) continuously reads frames and dispatches:RET-> pushes into the per-call Queue (synceval_r) or schedules async completion (+eval_r_async+)CALL-> spawns a new thread to execute the Ruby callback
- Callback threads send
RETback to R while keeping the reader thread free, which is required to support nested REQ servicing while waiting for a callback.
Failure behavior
- If the socket closes or a read error happens, pending REQ calls are signaled as closed and subsequent eval calls raise.
- Infrastructure failures (timeouts / broken pipes / connection reset) are
handled at a higher level by
RInstanceManagerrestart+retry logic.
Defined Under Namespace
Classes: Error, RProcessError, TimeoutError
Instance Attribute Summary collapse
-
#r_exit_status ⇒ Object
readonly
Returns the value of attribute r_exit_status.
-
#r_stderr ⇒ Object
readonly
Returns the value of attribute r_stderr.
Class Method Summary collapse
-
.callback_success_transport_payload(result) ⇒ Object
Gatekeeper parses RET success payload with strtod.
-
.default_use_unix?(bridge_target:) ⇒ Boolean
Whether to use Unix domain sockets for this client when
use_unixis nil.
Instance Method Summary collapse
-
#eval_r(code, session_id: 'default', instance_id: 'default', parent_id: nil, timeout: 60) ⇒ Hash
Evaluate one REQ on the connected R runtime and wait for the matching RET.
-
#eval_r_async(code, session_id: 'default', instance_id: 'default', parent_id: nil, timeout: nil, &block) ⇒ String
Schedule one REQ and invoke
blockon a background thread when RET arrives (or on failure). -
#initialize(source_path:, host: '127.0.0.1', bridge_host: nil, runtime_source_path: nil, r_cmd: 'R', use_unix: nil) ⇒ SessionClient
constructor
Create a SessionClient for one R runtime instance.
-
#register_callback(&block) ⇒ Object
Register a Ruby callback for R->Ruby
CALLenvelopes. -
#start(accept_timeout: 120) ⇒ Object
Start listening for a single TCP connection from one R process and launch that process.
-
#stop ⇒ Object
Stop the client and attempt to join threads cleanly.
Constructor Details
#initialize(source_path:, host: '127.0.0.1', bridge_host: nil, runtime_source_path: nil, r_cmd: 'R', use_unix: nil) ⇒ SessionClient
Create a SessionClient for one R runtime instance.
86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 |
# File 'lib/new_bridge/session_client.rb', line 86 def initialize(source_path:, host: '127.0.0.1', bridge_host: nil, runtime_source_path: nil, r_cmd: 'R', use_unix: nil) @source_path = File.(source_path) @use_precompiled = @source_path.end_with?('.so') @host = host @bridge_host = bridge_host || host @runtime_source_path = runtime_source_path || @source_path @r_cmd = r_cmd @use_unix = if use_unix.nil? self.class.default_use_unix?(bridge_target: @bridge_host) else use_unix end @unix_path = nil @server = nil @sock = nil @write_mx = Mutex.new @pending = {} @pending_mx = Mutex.new @callbacks = {} @callbacks_mx = Mutex.new @reader = nil @r_stderr = +'' @r_stderr_mx = Mutex.new @r_stderr_limit = 64 * 1024 @r_exit_status = nil end |
Instance Attribute Details
#r_exit_status ⇒ Object (readonly)
Returns the value of attribute r_exit_status.
74 75 76 |
# File 'lib/new_bridge/session_client.rb', line 74 def r_exit_status @r_exit_status end |
#r_stderr ⇒ Object (readonly)
Returns the value of attribute r_stderr.
74 75 76 |
# File 'lib/new_bridge/session_client.rb', line 74 def r_stderr @r_stderr end |
Class Method Details
.callback_success_transport_payload(result) ⇒ Object
Gatekeeper parses RET success payload with strtod. Contract:
register_callback_proc_stubends with nil → ACK "1" (semantic value staged in R).- Direct
register_callbackblocks may return a Numeric scalar; pass it through for legacy tests.
323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 |
# File 'lib/new_bridge/session_client.rb', line 323 def self.callback_success_transport_payload(result) case result when nil '1' when Numeric x = result.to_f (x.nan? || x.infinite?) ? '0' : x.to_s when true '1' when false '0' else '1' end end |
.default_use_unix?(bridge_target:) ⇒ Boolean
Whether to use Unix domain sockets for this client when use_unix is nil.
Policy:
GALAAZ_BRIDGE_TRANSPORT=tcp— always TCP (remote R, or debugging).GALAAZ_BRIDGE_TRANSPORT=unix— always Unix when supported.- Otherwise: Unix only if
bridge_targetis a local loopback address/name (127.0.0.1, ::1, localhost) or a Unix socket path (+#+). Any other host (e.g.host.docker.internal, LAN IP) uses TCP so R can reach Ruby over the network.
124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 |
# File 'lib/new_bridge/session_client.rb', line 124 def self.default_use_unix?(bridge_target:) return false unless defined?(UNIXServer) case ENV['GALAAZ_BRIDGE_TRANSPORT'].to_s.downcase when 'tcp' false when 'unix' true else t = bridge_target.to_s.strip if t.empty? false elsif t.start_with?('/') true else u = t.downcase %w[127.0.0.1 ::1 localhost].include?(u) end end end |
Instance Method Details
#eval_r(code, session_id: 'default', instance_id: 'default', parent_id: nil, timeout: 60) ⇒ Hash
Evaluate one REQ on the connected R runtime and wait for the matching RET.
244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 |
# File 'lib/new_bridge/session_client.rb', line 244 def eval_r(code, session_id: 'default', instance_id: 'default', parent_id: nil, timeout: 60) call_id = SecureRandom.uuid q = Queue.new @pending_mx.synchronize { @pending[call_id] = q } req = Envelope.encode( 'call_id' => call_id, 'type' => 'REQ', 'session_id' => session_id, 'instance_id' => instance_id, 'parent_id' => parent_id, 'payload' => code ) @write_mx.synchronize { Framing.write_frame(@sock, req) } ret = wait_for_ret(q, call_id, timeout) if ret['instance_id'] && ret['instance_id'] != instance_id raise RProcessError, "instance_id mismatch: expected=#{instance_id} got=#{ret['instance_id']}" end status = ret['status'] payload = ret['payload'] unless payload.is_a?(Hash) raise RProcessError, "invalid RET payload type=#{payload.class} (expected map)" end parsed = payload raise RProcessError, parsed['message'] || parsed.inspect unless status == 'success' parsed end |
#eval_r_async(code, session_id: 'default', instance_id: 'default', parent_id: nil, timeout: nil, &block) ⇒ String
Schedule one REQ and invoke block on a background thread when RET arrives (or on failure).
Unlike eval_r, this method returns immediately with the call_id String; it does not wait
for R unless timeout is nil (wait forever on the Ruby side for RET) or a Numeric (raise
completion with TimeoutError if RET is late).
282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 |
# File 'lib/new_bridge/session_client.rb', line 282 def eval_r_async(code, session_id: 'default', instance_id: 'default', parent_id: nil, timeout: nil, &block) raise ArgumentError, 'eval_r_async requires a block' unless block call_id = SecureRandom.uuid slot = AsyncPendingEntry.new(instance_id, block) @pending_mx.synchronize { @pending[call_id] = slot } req = Envelope.encode( 'call_id' => call_id, 'type' => 'REQ', 'session_id' => session_id, 'instance_id' => instance_id, 'parent_id' => parent_id, 'payload' => code ) @write_mx.synchronize { Framing.write_frame(@sock, req) } if timeout slot.timer_thread = Thread.new { async_timeout_wait(call_id, slot, timeout) } end call_id end |
#register_callback(&block) ⇒ Object
Register a Ruby callback for R->Ruby CALL envelopes.
The gatekeeper will invoke this callback when it receives a CALL with the
corresponding callback_call_id, and this client will reply with a RET.
The same call_id may be used many times (R stub function is reused); the block
stays registered until the client stops — do not delete on first CALL.
312 313 314 315 316 |
# File 'lib/new_bridge/session_client.rb', line 312 def register_callback(&block) callback_call_id = SecureRandom.uuid @callbacks_mx.synchronize { @callbacks[callback_call_id] = block } callback_call_id end |
#start(accept_timeout: 120) ⇒ Object
Start listening for a single TCP connection from one R process and launch that process.
- Binds to an ephemeral TCP port, or a Unix socket path when
use_unixis set. - Starts a thread that runs
r_cmdand sources the gatekeeper C++ code inside R. - Accepts the runtime connection with
accept_timeout.
On startup failure (e.g. runtime exits before connecting), it surfaces R stderr and exit status.
152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 |
# File 'lib/new_bridge/session_client.rb', line 152 def start(accept_timeout: 120) use_u = @use_unix && defined?(UNIXServer) if @use_unix && !defined?(UNIXServer) warn '[NewBridge::SessionClient] Unix bridge transport selected but UNIXServer is unavailable; falling back to TCP' use_u = false end if use_u @unix_path = File.join(Dir.tmpdir, "galaaz_bridge_#{Process.pid}_#{SecureRandom.hex(6)}.sock") File.unlink(@unix_path) if File.exist?(@unix_path) @server = UNIXServer.new(@unix_path) port = 0 bridge_for_r = @unix_path.gsub("'", "\\\\'") else @server = TCPServer.new(@host, 0) port = @server.addr[1] bridge_for_r = @bridge_host.gsub("'", "\\\\'") end # Use sourceCpp with a persistent cache directory so repeated runtime # starts can reuse compiled shared objects when source is unchanged. cpp_escaped = @runtime_source_path.gsub("'", "\\\\'") rcpp_cache_dir = (ENV['GALAAZ_RCPP_CACHE_DIR'] || File.join(Dir.tmpdir, 'galaaz_rcpp_cache')).gsub("'", "\\\\'") r_script = <<~R bridge_host <- '#{bridge_for_r}' port <- #{port}L stopifnot(requireNamespace("Rcpp", quietly = TRUE)) library(Rcpp) dir.create("#{rcpp_cache_dir}", recursive = TRUE, showWarnings = FALSE) sourceCpp(file = "#{cpp_escaped}", rebuild = FALSE, cacheDir = "#{rcpp_cache_dir}") galaaz_run_bridge(bridge_host, as.integer(port)) R env = if use_u { 'GALAAZ_BRIDGE_HOST' => @unix_path, 'GALAAZ_BRIDGE_PORT' => '0', 'GALAAZ_BRIDGE_UNIX' => '1' } else { 'GALAAZ_BRIDGE_HOST' => @bridge_host, 'GALAAZ_BRIDGE_PORT' => port.to_s } end @r_thr = Thread.new do launch = @r_cmd.is_a?(Array) ? @r_cmd.dup : [@r_cmd] _stdin, stdout_err, wait_thr = Open3.popen2e(env, *launch, '--slave', '--no-save', '-e', r_script) # When the R process exits or during shutdown, the underlying pipe can # close while this background thread is still blocked reading. # Treat that as a normal shutdown and avoid JRuby "stream closed in # another thread" warnings. begin stdout_err.each_line { |line| append_r_stderr(line) } rescue IOError, Errno::EPIPE, Errno::ECONNRESET # normal shutdown / pipe closure end @r_exit_status = wait_thr.value end @sock = Timeout.timeout(accept_timeout) { @server.accept } @reader = Thread.new { reader_loop } self rescue Timeout::Error # If the runtime process failed early (e.g. docker permission/network/source path), # surface stderr/exit details instead of a generic accept timeout. process_status = @r_exit_status&.exitstatus details = current_r_stderr msg = "failed to accept runtime connection within #{accept_timeout}s" msg += " (runtime exit=#{process_status})" if process_status unless details.empty? tail = details.length > 2000 ? details[-2000, 2000] : details msg += " stderr_tail=#{tail}" end raise RProcessError, msg end |
#stop ⇒ Object
Stop the client and attempt to join threads cleanly. Best-effort: socket/server are closed, reader thread is joined, runtime thread is joined.
224 225 226 227 228 229 230 231 232 233 |
# File 'lib/new_bridge/session_client.rb', line 224 def stop @sock&.close rescue nil @server&.close rescue nil if @unix_path && File.socket?(@unix_path) File.unlink(@unix_path) rescue nil end @unix_path = nil @reader&.join(3) @r_thr&.join(15) end |