Concurrent requests overwrite each other's instrumentation data

Abierto
  • ¿Algo no está bien?
#657 0 comentarios 0 reacciones 0 asignados Ver en GitHub

Los mantenedores suelen responder en 1 día

@atesgoral ya está trabajando en esto.

Desde el 6/11/2026.

  • #668 de @atesgoral — abierto

Evaluación

Dificultad
3/5
Tiempo estimado
1-2 días
Aptitud para principiantes
35/100
Tipo de issue
Error
Claridad
Bien especificado
Estado de actividad
Estancado
Stack tecnológico
ruby
Área
backend, observability

Línea de trabajo

Start with lib/mcp/instrumentation.rb, especially instrument_call and add_instrumentation_data, then read the concurrent shared server test in test/mcp/server_test.rb and the tool call instrumentation in lib/mcp/server.rb. Reproduce the interleaving described in the issue and check how request data reaches the callback; done when concurrent requests produce separate records with the correct tool names. An open pull request (#668) is already addressing this issue.

Escrito por el modelo de indexación a partir del texto del issue.

Descripción

bug

Instrumentation#instrument_call keeps the per-request record in @instrumentation_data, an instance variable on the shared Server (instrumentation.rb:8). When two requests run at the same time on one server, for example Streamable HTTP on a threaded Rack server, the second request replaces the hash.

The first request then writes its later fields (client, error, duration) into the second request's hash and passes that hash to instrumentation_callback (instrumentation.rb:43, :48). The first request's own record, with its tool_name and tool_arguments (server.rb:1408), is never sent to the callback. around_request gets the original hash (instrumentation.rb:17) but misses the fields written after the swap.

Concurrent use of one server is supported (see test "#handle concurrent requests on a shared server report exceptions independently" in test/mcp/server_test.rb).

Repro

Run from the repository root with bundle exec ruby -I lib repro.rb:

require "mcp"

entered = Queue.new
release = Queue.new
slow_tool = MCP::Tool.define(name: "slow_tool") do
  entered << true
  release.pop
  MCP::Tool::Response.new([{ type: "text", text: "slow" }])
end
fast_tool = MCP::Tool.define(name: "fast_tool") do
  MCP::Tool::Response.new([{ type: "text", text: "fast" }])
end

records = Queue.new
configuration = MCP::Configuration.new
configuration.instrumentation_callback = ->(data) { records << data.dup }
server = MCP::Server.new(name: "example", tools: [slow_tool, fast_tool], configuration: configuration)

call = ->(id, name) { server.handle({ jsonrpc: "2.0", id: id, method: "tools/call", params: { name: name, arguments: {} } }) }
thread = Thread.new { call.call(1, "slow_tool") }
entered.pop
call.call(2, "fast_tool")
release << true
thread.join

p(2.times.map { records.pop[:tool_name] })

Expected: ["fast_tool", "slow_tool"]

Actual: ["fast_tool", "fast_tool"]

Suggested fix

Keep the record local to the request instead of on the shared server. For example, create the hash in instrument_call and pass it to add_instrumentation_data, or keep it in a fiber-local value that instrument_call sets and restores.

Lenguaje dominante
Ruby
Estrellas
940
Forks
143
Merge medio
1 d 5 h
PR fusionados (30 d)
48

Preparar el entorno

Primeros pasos

  1. Lee el issue completo y luego la guía de contribución del proyecto.
  2. Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
  3. Haz un fork del repositorio y trabaja en una rama.
  4. Abre un pull request que haga referencia al número del issue.

Más de modelcontextprotocol/ruby-sdk

Todos los issues de modelcontextprotocol/ruby-sdk

Issues similares

Más issues de Ruby

Recibe los nuevos issues en tu correo

Un resumen breve de issues de GitHub para principiantes.