feat: add retry logic and health check for server connection

Author: sudo-tee

lua/opencode/config_file.lua

@@ -10,13 +10,16 @@ local M = {
 M.get_opencode_config = Promise.async(function()
   if not M.config_promise then
     local state = require('opencode.state')
-    M.config_promise = state.api_client:get_config()
+    M.config_promise = Promise.retry(function()
+      return state.api_client:get_config()
+    end, 3, 500)
   end
   local ok, result = pcall(function()
     return M.config_promise:await()
   end)
 
   if not ok then
+    M.config_promise = nil
     vim.notify('Error fetching Opencode config: ' .. vim.inspect(result), vim.log.levels.ERROR)
     return nil
   end
@@ -28,12 +31,15 @@ end)
 M.get_opencode_project = Promise.async(function()
   if not M.project_promise then
     local state = require('opencode.state')
-    M.project_promise = state.api_client:get_current_project()
+    M.project_promise = Promise.retry(function()
+      return state.api_client:get_current_project()
+    end, 3, 500)
   end
   local ok, result = pcall(function()
     return M.project_promise:await()
   end)
   if not ok then
+    M.project_promise = nil
     vim.notify('Error fetching Opencode project: ' .. vim.inspect(result), vim.log.levels.ERROR)
     return nil
   end
@@ -77,6 +83,7 @@ function M.get_opencode_providers()
   end
   local wrapped = M.providers_promise:catch(function(err)
     vim.notify('Error fetching Opencode providers: ' .. vim.inspect(err), vim.log.levels.ERROR)
+    M.providers_promise = nil
     return nil
   end)
   if not _providers_render_callback then

lua/opencode/git_review.lua

@@ -286,9 +286,7 @@ M.revert_selected_file = require_git_project(function(ref)
 end)
 
 M.revert_all = require_git_project(function(ref)
-  vim.print('⭕ ❱ git_review.lua:288 ❱ ƒ(anonymous) ❱ ref =', ref)
   M.__current_ref = ref or M.get_first_snapshot()
-  vim.print('⭕ ❱ git_review.lua:289 ❱ ƒ(M.__current_ref) ❱ M.__current_ref =', M.__current_ref)
 
   local files = get_changed_files()
 

lua/opencode/opencode_server.lua

@@ -86,6 +86,37 @@ function OpencodeServer:is_running()
   return self.job.pid ~= nil
 end
 
+---Perform a health check on a server URL.
+---@param url string The full health endpoint URL
+---@param timeout_ms number Timeout in milliseconds
+---@return Promise<boolean>
+function OpencodeServer.health_check(url, timeout_ms)
+  local health_promise = Promise.new()
+  curl.request({
+    url = url,
+    method = 'GET',
+    timeout = timeout_ms or 2000,
+    proxy = '',
+    callback = function(response)
+      health_promise:resolve(response ~= nil and response.status >= 200 and response.status < 300)
+    end,
+    on_error = function(_err)
+      health_promise:resolve(false)
+    end,
+  })
+  return health_promise
+end
+
+---Check if the server is reachable via its health endpoint.
+---@return Promise<boolean>
+function OpencodeServer:check_health()
+  if not self.url then
+    return Promise.new():resolve(false)
+  end
+  local health_url = self.url:gsub('/$', '') .. '/global/health'
+  return OpencodeServer.health_check(health_url, 2000)
+end
+
 local function kill_process(pid, signal, desc)
   local log = require('opencode.log')
   local ok, err = pcall(vim.uv.kill, pid, signal)

lua/opencode/promise.lua

@@ -418,4 +418,40 @@ function Promise.all(promises)
   end)
 end
 
+---Create a promise that resolves after a delay.
+---@param ms number Delay in milliseconds
+---@return Promise<boolean>
+function Promise.delay(ms)
+  local delay = Promise.new()
+  vim.defer_fn(function()
+    pcall(delay.resolve, delay, true)
+  end, ms)
+  return delay
+end
+
+---Retry a promise-returning operation on failure.
+---@generic T
+---@param factory fun(): Promise<T> Creates a fresh promise per attempt
+---@param max_retries number Total attempts (1 = no retry)
+---@param delay_ms number Delay between retries in milliseconds
+---@return Promise<T>
+function Promise.retry(factory, max_retries, delay_ms)
+  return Promise.spawn(function()
+    local last_err
+    for i = 1, max_retries do
+      local ok, result = pcall(function()
+        return factory():await()
+      end)
+      if ok then
+        return result
+      end
+      last_err = result
+      if i < max_retries then
+        Promise.delay(delay_ms):await()
+      end
+    end
+    return Promise.new():reject(last_err)
+  end)
+end
+
 return Promise

lua/opencode/server_job.lua

@@ -20,38 +20,20 @@ end
 --- @param timeout number
 --- @return Promise<string|nil>
 local function try_custom_server(base_url, timeout)
-  local promise = Promise.new()
   local health_url = base_url .. '/global/health'
 
   log.debug('try_custom_server: checking health at %s', health_url)
 
-  curl.request({
-    url = health_url,
-    method = 'GET',
-    timeout = timeout * 1000,
-    proxy = '', -- Disable proxy for health check
-    callback = function(response)
-      if response and response.status >= 200 and response.status < 300 then
-        local success, health_data = pcall(vim.json.decode, response.body)
-        if success and health_data then
-          log.debug('try_custom_server: health check passed')
-          promise:resolve(base_url)
-          return
-        end
-      end
-
-      local err_msg =
-        string.format('Health check failed at %s (status: %d)', health_url, response and response.status or 0)
-      log.debug('try_custom_server: %s', err_msg)
-      promise:reject(err_msg)
-    end,
-    on_error = function(err)
-      log.debug('try_custom_server: error connecting to %s: %s', health_url, vim.inspect(err))
-      promise:reject(err)
-    end,
-  })
+  return opencode_server.health_check(health_url, timeout * 1000):and_then(function(healthy)
+    if healthy then
+      log.debug('try_custom_server: health check passed')
+      return base_url
+    end
 
-  return promise
+    local err_msg = string.format('Health check failed at %s', health_url)
+    log.debug('try_custom_server: %s', err_msg)
+    return Promise.new():reject(err_msg)
+  end)
 end
 
 --- @param response {status: integer, body: string}
@@ -185,15 +167,9 @@ local function resolve_port()
   return existing or math.random(1024, 65535)
 end
 
---- Ensure the opencode server is running, starting it if necessary.
---- @return Promise<OpencodeServer>
-function M.ensure_server()
+local function _start_server()
   local promise = Promise.new()
 
-  if state.opencode_server and state.opencode_server:is_running() then
-    return promise:resolve(state.opencode_server)
-  end
-
   local custom_url = config.server.url
   if not custom_url then
     log.debug('ensure_server: server.url not configured, spawning local server')
@@ -218,20 +194,36 @@ function M.ensure_server()
   return promise
 end
 
-local function retry_connect(base_url, timeout, max_retries, on_success, on_failure)
-  local function attempt(retry_count)
-    vim.defer_fn(function()
-      try_custom_server(base_url, timeout):and_then(on_success):catch(function(err)
-        if retry_count < max_retries then
-          attempt(retry_count + 1)
-        else
-          log.error('try_connect_to_custom_server: exhausted %d retries: %s', max_retries, vim.inspect(err))
-          on_failure(err)
-        end
-      end)
-    end, retry_count * (config.server.retry_delay or 2000))
+--- Ensure the opencode server is running, starting it if necessary.
+--- @return Promise<OpencodeServer>
+function M.ensure_server()
+  if state.opencode_server and state.opencode_server:is_running() then
+    return state.opencode_server:check_health():and_then(function(healthy)
+      if healthy then
+        return state.opencode_server
+      end
+      log.warn('ensure_server: cached server unhealthy, reconnecting')
+      state.jobs.clear_server()
+      return _start_server()
+    end)
   end
-  attempt(1)
+
+  return _start_server()
+end
+
+local function retry_connect(base_url, timeout, max_retries, on_success, on_failure)
+  local delay = config.server.retry_delay or 2000
+  Promise.delay(delay)
+    :and_then(function()
+      return Promise.retry(function()
+        return try_custom_server(base_url, timeout)
+      end, max_retries, delay)
+    end)
+    :and_then(on_success)
+    :catch(function(err)
+      log.error('try_connect_to_custom_server: exhausted %d retries: %s', max_retries, vim.inspect(err))
+      on_failure(err)
+    end)
 end
 
 local function spawn_and_retry(base_url, custom_port, custom_url, promise, timeout)

tests/unit/persist_state_spec.lua

@@ -197,6 +197,9 @@ describe('persist_state', function()
       is_running = function()
         return true
       end,
+      check_health = function()
+        return Promise.new():resolve(true)
+      end,
       spawn = function() end,
       shutdown = function()
         return Promise.new():resolve(true)

tests/unit/server_job_spec.lua

@@ -1,4 +1,5 @@
 local server_job = require('opencode.server_job')
+local Promise = require('opencode.promise')
 
 local curl = require('opencode.curl')
 local assert = require('luassert')
@@ -93,6 +94,9 @@ describe('server_job', function()
         end)
       end,
       shutdown = function() end,
+      check_health = function()
+        return Promise.new():resolve(true)
+      end,
     }
     opencode_server.new = function()
       return fake

tests/unit/services_session_runtime_spec.lua

@@ -85,6 +85,9 @@ describe('opencode.services.session_runtime', function()
       end,
       shutdown = function() end,
       url = 'http://127.0.0.1:4000',
+      check_health = function()
+        return Promise.new():resolve(true)
+      end,
     })
   end)