Tensorflow Serving 源码阅读笔记

Tensorflow Serving 源码阅读笔记

模型加载过程

由 tensorflow_serving/model_servers/BUILD 配置

341 cc_library(
342     name = "tensorflow_model_server_main_lib",
343     srcs = [
344         "main.cc",
345     ],
346     hdrs = [
347         "version.h",
348     ],
349     linkstamp = "version.cc",
350     visibility = [
351         ":tensorflow_model_server_custom_op_clients",
352         "//tensorflow_serving:internal",
353     ],
354     deps = [
355         ":server_lib",
356         "@org_tensorflow//tensorflow/c:c_api",
357         "@org_tensorflow//tensorflow/core:lib",
358         "@org_tensorflow//tensorflow/core/platform/cloud:gcs_file_system",
359         "@org_tensorflow//tensorflow/core/platform/hadoop:hadoop_file_system",
360         "@org_tensorflow//tensorflow/core/platform/s3:s3_file_system",
361     ],
362 )

364 cc_binary(
365     name = "tensorflow_model_server",
366     stamp = 1,
367     visibility = [
368         ":testing",
369         "//tensorflow_serving:internal",
370     ],
371     linkopts = ["-rdynamic"],
372     deps = [
373         ":tensorflow_model_server_main_lib",
374         "@org_tensorflow//tensorflow/c:c_api",
375     ],
376 )

可知 tensorflow_model_server 的入口在 tensorflow_serving/model_servers/main.cc，参数 model_base_path 指定模型的路径：

 94       tensorflow::Flag("model_base_path", &options.model_base_path,
 95                        "path to export (i