AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Implement the APIs for gRPC calls in client/server

Browse Source 
Implement the APIs defined in the RdcStandaloneHandler to make gRPC call to daemon

Implement the APIs defined in the RdcAPIServiceImpl to handle the gRPC calls in daemon

Add two APIs to get all GPU groups and field groups: rdc_group_get_all_ids()
and rdc_group_field_all_ids()
Those two APIs are required by the rdci group and fieldgroup
sub-modules.

Change-Id: I066091423146dea180c16af212688ed43dc44611


[ROCm/rdc commit: 7ee29b6cdd]
This commit is contained in:
Bill(Shuzhou) Liu
2020-03-18 08:58:48 -04:00
committed by Chris Freehill
parent 3d8180e5af
commit ce4890f88c
23 changed files with 907 additions and 154 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rdc/include/rdc/rdc.h
+57 -14
View File
@@ -59,7 +59,7 @@ typedef enum {
* @brief rdc operation mode
* rdc can run in auto mode where background threads will collect metrics.
* When run in manual mode, the user needs to periodically call
* rdc_update_all_fields for data collection.
* rdc_field_update_all for data collection.
*/
typedef enum {
RDC_OPERATION_MODE_AUTO = 0,
@@ -109,6 +109,15 @@ typedef enum {
*/
#define RDC_MAX_FIELD_IDS_PER_FIELD_GROUP 128
/**
* @brief The max number of groups
*/
#define RDC_MAX_NUM_GROUPS 64
/**
* @brief The max number of the field groups
*/
#define RDC_MAX_NUM_FIELD_GROUPS 64
/**
* @biref The fields
@@ -279,12 +288,12 @@ rdc_status_t rdc_shutdown();
*
* @details The RDC is loaded as library so that it does not require rdcd
* daemon. In this mode, the user has to periodically call
* rdc_update_all_fields() when op_mode is RDC_OPERATION_MODE_MANUAL, which
* rdc_field_update_all() when op_mode is RDC_OPERATION_MODE_MANUAL, which
* tells RDC to collect the stats. This function is not thread safe.
*
* @param[in] op_mode Operation modes. When RDC_OPERATION_MODE_AUTO, RDC schedules
* background task to collect the stats. When RDC_OPERATION_MODE_MANUAL, the user
* needs to call rdc_update_all_fields() periodically.
* needs to call rdc_field_update_all() periodically.
*
* @param[inout] p_rdc_handle Caller provided pointer to rdc_handle_t. Upon
* successful call, the value will contain the handler for following API calls.
@@ -351,7 +360,7 @@ rdc_status_t rdc_disconnect(rdc_handle_t p_rdc_handle);
*
* @details This should be executed as part of job prologue. The summary
* job stats can be retrieved using rdc_job_get_stats().
* In RDC_OPERATION_MODE_MANUAL, user must call rdc_update_all_fields(1)
* In RDC_OPERATION_MODE_MANUAL, user must call rdc_field_update_all(1)
* at least once, before call rdc_job_get_stats()
*
* @param[in] p_rdc_handle The RDC handler.
@@ -419,7 +428,7 @@ rdc_status_t rdc_job_stop_stats(rdc_handle_t p_rdc_handle,
*
* @retval ::RDC_ST_OK is returned upon successful call.
*/
rdc_status_t rdc_update_all_fields(rdc_handle_t p_rdc_handle,
rdc_status_t rdc_field_update_all(rdc_handle_t p_rdc_handle,
uint32_t wait_for_update);
/**
@@ -438,7 +447,7 @@ rdc_status_t rdc_update_all_fields(rdc_handle_t p_rdc_handle,
*
* @retval ::RDC_ST_OK is returned upon successful call.
*/
rdc_status_t rdc_get_all_devices(rdc_handle_t p_rdc_handle,
rdc_status_t rdc_device_get_all(rdc_handle_t p_rdc_handle,
uint32_t gpu_index_list[RDC_MAX_NUM_DEVICES], uint32_t* count);
/**
@@ -455,7 +464,7 @@ rdc_status_t rdc_get_all_devices(rdc_handle_t p_rdc_handle,
*
* @retval ::RDC_ST_OK is returned upon successful call.
*/
rdc_status_t rdc_get_device_attributes(rdc_handle_t p_rdc_handle,
rdc_status_t rdc_device_get_attributes(rdc_handle_t p_rdc_handle,
uint32_t gpu_index, rdc_device_attributes_t* p_rdc_attr);
/**
@@ -518,6 +527,23 @@ rdc_status_t rdc_group_gpu_add(rdc_handle_t p_rdc_handle,
rdc_status_t rdc_group_gpu_get_info(rdc_handle_t p_rdc_handle,
rdc_gpu_group_t p_rdc_group_id, rdc_group_info_t* p_rdc_group_info);
/**
* @brief Used to get information about all GPU groups in the system.
*
* @details Get the list of GPU group ids in the system.
*
* @param[in] p_rdc_handle The RDC handler.
*
* @param[out] group_id_list Array reference to fill GPU group
* ids in the system.
*
* @param[out] count Number of GPU group returned in group_id_list.
*
* @retval ::RDC_ST_OK is returned upon successful call.
*/
rdc_status_t rdc_group_get_all_ids(rdc_handle_t p_rdc_handle,
rdc_gpu_group_t group_id_list[], uint32_t* count);
/**
* @brief Destroy GPU group represented by p_rdc_group_id
*
@@ -576,6 +602,23 @@ rdc_status_t rdc_group_field_get_info(rdc_handle_t p_rdc_handle,
rdc_field_grp_t rdc_field_group_id,
rdc_field_group_info_t* field_group_info);
/**
* @brief Used to get information about all field groups in the system.
*
* @details Get the list of field group ids in the system.
*
* @param[in] p_rdc_handle The RDC handler.
*
* @param[out] field_group_id_list Array reference to fill field group
* ids in the system.
*
* @param[out] count Number of field group returned in field_group_id_list.
*
* @retval ::RDC_ST_OK is returned upon successful call.
*/
rdc_status_t rdc_group_field_get_all_ids(rdc_handle_t p_rdc_handle,
rdc_field_grp_t field_group_id_list[], uint32_t* count);
/**
* @brief Destroy field group represented by rdc_field_group_id
*
@@ -596,7 +639,7 @@ rdc_status_t rdc_group_field_destroy(rdc_handle_t p_rdc_handle,
*
* @details Note that the first update of the field will not occur
* until the next field update cycle. To force a field update cycle,
* user must call rdc_update_all_fields(1)
* user must call rdc_field_update_all(1)
*
* @param[in] p_rdc_handle The RDC handler.
*
@@ -612,14 +655,14 @@ rdc_status_t rdc_group_field_destroy(rdc_handle_t p_rdc_handle,
*
* @retval ::RDC_ST_OK is returned upon successful call.
*/
rdc_status_t rdc_watch_fields(rdc_handle_t p_rdc_handle,
rdc_status_t rdc_field_watch(rdc_handle_t p_rdc_handle,
rdc_gpu_group_t group_id, rdc_field_grp_t field_group_id,
uint64_t update_freq, double max_keep_age, uint32_t max_keep_samples);
/**
* @brief Request a latest cached field of a GPU
*
* @details Note that the field can be cached after called rdc_watch_fields
* @details Note that the field can be cached after called rdc_field_watch
*
* @param[in] p_rdc_handle The RDC handler.
*
@@ -631,13 +674,13 @@ rdc_status_t rdc_watch_fields(rdc_handle_t p_rdc_handle,
*
* @retval ::RDC_ST_OK is returned upon successful call.
*/
rdc_status_t rdc_get_latest_value_for_field(rdc_handle_t p_rdc_handle,
rdc_status_t rdc_field_get_latest_value(rdc_handle_t p_rdc_handle,
uint32_t gpu_index, uint32_t field, rdc_field_value* value);
/**
* @brief Request a history cached field of a GPU
*
* @details Note that the field can be cached after called rdc_watch_fields
* @details Note that the field can be cached after called rdc_field_watch
*
* @param[in] p_rdc_handle The RDC handler.
*
@@ -655,7 +698,7 @@ rdc_status_t rdc_get_latest_value_for_field(rdc_handle_t p_rdc_handle,
*
* @retval ::RDC_ST_OK is returned upon successful call.
*/
rdc_status_t rdc_get_field_value_since(rdc_handle_t p_rdc_handle,
rdc_status_t rdc_field_get_value_since(rdc_handle_t p_rdc_handle,
uint32_t gpu_index, uint32_t field, uint64_t since_time_stamp,
uint64_t *next_since_time_stamp, rdc_field_value* value);
@@ -672,7 +715,7 @@ rdc_status_t rdc_get_field_value_since(rdc_handle_t p_rdc_handle,
*
* @retval ::RDC_ST_OK is returned upon successful call.
*/
rdc_status_t rdc_unwatch_fields(rdc_handle_t p_rdc_handle,
rdc_status_t rdc_field_unwatch(rdc_handle_t p_rdc_handle,
rdc_gpu_group_t group_id, rdc_field_grp_t field_group_id);
/**
projects/rdc/include/rdc_lib/RdcCacheManager.h
+2 -2
View File
@@ -34,9 +34,9 @@ namespace rdc {
class RdcCacheManager {
public:
virtual rdc_status_t rdc_get_latest_value_for_field(uint32_t gpu_index,
virtual rdc_status_t rdc_field_get_latest_value(uint32_t gpu_index,
uint32_t field, rdc_field_value* value) = 0;
virtual rdc_status_t rdc_get_field_value_since(uint32_t gpu_index,
virtual rdc_status_t rdc_field_get_value_since(uint32_t gpu_index,
uint32_t field, uint64_t since_time_stamp,
uint64_t *next_since_time_stamp, rdc_field_value* value) = 0;
virtual rdc_status_t rdc_update_cache(uint32_t gpu_index,
projects/rdc/include/rdc_lib/RdcGroupSettings.h
+5
View File
@@ -40,6 +40,9 @@ class RdcGroupSettings {
virtual rdc_status_t rdc_group_gpu_get_info(
rdc_gpu_group_t p_rdc_group_id,
rdc_group_info_t* p_rdc_group_info) = 0;
virtual rdc_status_t rdc_group_get_all_ids(
rdc_gpu_group_t group_id_list[], uint32_t* count) = 0;
virtual rdc_status_t rdc_group_field_create(uint32_t num_field_ids,
uint32_t* field_ids, const char* field_group_name,
@@ -49,6 +52,8 @@ class RdcGroupSettings {
virtual rdc_status_t rdc_group_field_get_info(
rdc_field_grp_t rdc_field_group_id,
rdc_field_group_info_t* field_group_info) = 0;
virtual rdc_status_t rdc_group_field_get_all_ids(
rdc_field_grp_t field_group_id_list[], uint32_t* count) = 0;
virtual ~RdcGroupSettings() {}
};
projects/rdc/include/rdc_lib/RdcHandler.h
+11 -7
View File
@@ -41,9 +41,9 @@ class RdcHandler {
// Discovery API
virtual rdc_status_t rdc_get_all_devices(
virtual rdc_status_t rdc_device_get_all(
uint32_t gpu_index_list[RDC_MAX_NUM_DEVICES], uint32_t* count) = 0;
virtual rdc_status_t rdc_get_device_attributes(uint32_t gpu_index,
virtual rdc_status_t rdc_device_get_attributes(uint32_t gpu_index,
rdc_device_attributes_t* p_rdc_attr) = 0;
// Group API
@@ -59,25 +59,29 @@ class RdcHandler {
rdc_field_group_info_t* field_group_info) = 0;
virtual rdc_status_t rdc_group_gpu_get_info(
rdc_gpu_group_t p_rdc_group_id, rdc_group_info_t* p_rdc_group_info) = 0;
virtual rdc_status_t rdc_group_get_all_ids(
rdc_gpu_group_t group_id_list[], uint32_t* count) = 0;
virtual rdc_status_t rdc_group_field_get_all_ids(
rdc_field_grp_t field_group_id_list[], uint32_t* count) = 0;
virtual rdc_status_t rdc_group_gpu_destroy(
rdc_gpu_group_t p_rdc_group_id) = 0;
virtual rdc_status_t rdc_group_field_destroy(
rdc_field_grp_t rdc_field_group_id) = 0;
// Field API
virtual rdc_status_t rdc_watch_fields(rdc_gpu_group_t group_id,
virtual rdc_status_t rdc_field_watch(rdc_gpu_group_t group_id,
rdc_field_grp_t field_group_id, uint64_t update_freq,
double max_keep_age, uint32_t max_keep_samples) = 0;
virtual rdc_status_t rdc_get_latest_value_for_field(uint32_t gpu_index,
virtual rdc_status_t rdc_field_get_latest_value(uint32_t gpu_index,
uint32_t field, rdc_field_value* value) = 0;
virtual rdc_status_t rdc_get_field_value_since(uint32_t gpu_index,
virtual rdc_status_t rdc_field_get_value_since(uint32_t gpu_index,
uint32_t field, uint64_t since_time_stamp,
uint64_t *next_since_time_stamp, rdc_field_value* value) = 0;
virtual rdc_status_t rdc_unwatch_fields(rdc_gpu_group_t group_id,
virtual rdc_status_t rdc_field_unwatch(rdc_gpu_group_t group_id,
rdc_field_grp_t field_group_id) = 0;
// Control API
virtual rdc_status_t rdc_update_all_fields(uint32_t wait_for_update) = 0;
virtual rdc_status_t rdc_field_update_all(uint32_t wait_for_update) = 0;
virtual ~RdcHandler(){}
};
projects/rdc/include/rdc_lib/RdcWatchTable.h
+3 -3
View File
@@ -33,7 +33,7 @@ namespace rdc {
class RdcWatchTable {
public:
virtual rdc_status_t rdc_update_all_fields() = 0;
virtual rdc_status_t rdc_field_update_all() = 0;
virtual rdc_status_t rdc_job_start_stats(rdc_gpu_group_t group_id,
char job_id[64]) = 0;
@@ -41,10 +41,10 @@ class RdcWatchTable {
uint64_t update_freq, double max_keep_age,
uint32_t max_keep_samples) = 0;
virtual rdc_status_t rdc_watch_fields(rdc_gpu_group_t group_id,
virtual rdc_status_t rdc_field_watch(rdc_gpu_group_t group_id,
rdc_field_grp_t field_group_id, uint64_t update_freq,
double max_keep_age, uint32_t max_keep_samples) = 0;
virtual rdc_status_t rdc_unwatch_fields(rdc_gpu_group_t group_id,
virtual rdc_status_t rdc_field_unwatch(rdc_gpu_group_t group_id,
rdc_field_grp_t field_group_id) = 0;
virtual ~RdcWatchTable() {}
projects/rdc/include/rdc_lib/impl/RdcCacheManagerImpl.h
+2 -2
View File
@@ -44,9 +44,9 @@ typedef std::map<RdcFieldKey, std::vector<RdcCacheEntry>> RdcCacheSamples;
class RdcCacheManagerImpl: public RdcCacheManager {
public:
rdc_status_t rdc_get_latest_value_for_field(uint32_t gpu_index,
rdc_status_t rdc_field_get_latest_value(uint32_t gpu_index,
uint32_t field, rdc_field_value* value) override;
rdc_status_t rdc_get_field_value_since(uint32_t gpu_index, uint32_t field,
rdc_status_t rdc_field_get_value_since(uint32_t gpu_index, uint32_t field,
uint64_t since_time_stamp, uint64_t *next_since_time_stamp,
rdc_field_value* value) override;
rdc_status_t rdc_update_cache(uint32_t gpu_index,
projects/rdc/include/rdc_lib/impl/RdcEmbeddedHandler.h
+11 -7
View File
@@ -43,9 +43,9 @@ class RdcEmbeddedHandler: public RdcHandler {
rdc_status_t rdc_job_stop_stats(char job_id[64]) override;
// Discovery API
rdc_status_t rdc_get_all_devices(
rdc_status_t rdc_device_get_all(
uint32_t gpu_index_list[RDC_MAX_NUM_DEVICES], uint32_t* count) override;
rdc_status_t rdc_get_device_attributes(uint32_t gpu_index,
rdc_status_t rdc_device_get_attributes(uint32_t gpu_index,
rdc_device_attributes_t* p_rdc_attr) override;
// Group API
@@ -62,25 +62,29 @@ class RdcEmbeddedHandler: public RdcHandler {
rdc_field_group_info_t* field_group_info) override;
rdc_status_t rdc_group_gpu_get_info(rdc_gpu_group_t p_rdc_group_id,
rdc_group_info_t* p_rdc_group_info) override;
rdc_status_t rdc_group_get_all_ids(
rdc_gpu_group_t group_id_list[], uint32_t* count) override;
rdc_status_t rdc_group_field_get_all_ids(
rdc_field_grp_t field_group_id_list[], uint32_t* count) override;
rdc_status_t rdc_group_gpu_destroy(
rdc_gpu_group_t p_rdc_group_id) override;
rdc_status_t rdc_group_field_destroy(
rdc_field_grp_t rdc_field_group_id) override;
// Field API
rdc_status_t rdc_watch_fields(rdc_gpu_group_t group_id,
rdc_status_t rdc_field_watch(rdc_gpu_group_t group_id,
rdc_field_grp_t field_group_id, uint64_t update_freq,
double max_keep_age, uint32_t max_keep_samples) override;
rdc_status_t rdc_get_latest_value_for_field(uint32_t gpu_index,
rdc_status_t rdc_field_get_latest_value(uint32_t gpu_index,
uint32_t field, rdc_field_value* value) override;
rdc_status_t rdc_get_field_value_since(uint32_t gpu_index,
rdc_status_t rdc_field_get_value_since(uint32_t gpu_index,
uint32_t field, uint64_t since_time_stamp,
uint64_t *next_since_time_stamp, rdc_field_value* value) override;
rdc_status_t rdc_unwatch_fields(rdc_gpu_group_t group_id,
rdc_status_t rdc_field_unwatch(rdc_gpu_group_t group_id,
rdc_field_grp_t field_group_id) override;
// Control API
rdc_status_t rdc_update_all_fields(uint32_t wait_for_update) override;
rdc_status_t rdc_field_update_all(uint32_t wait_for_update) override;
explicit RdcEmbeddedHandler(rdc_operation_mode_t op_mode);
~RdcEmbeddedHandler();
projects/rdc/include/rdc_lib/impl/RdcGroupSettingsImpl.h
+5 -1
View File
@@ -25,7 +25,7 @@ THE SOFTWARE.
#include <memory>
#include <map>
#include <mutex>
#include <mutex> // NOLINT
#include <string>
#include "rdc_lib/RdcGroupSettings.h"
@@ -42,6 +42,8 @@ class RdcGroupSettingsImpl: public RdcGroupSettings {
uint32_t gpu_index) override;
rdc_status_t rdc_group_gpu_get_info(rdc_gpu_group_t p_rdc_group_id,
rdc_group_info_t* p_rdc_group_info) override;
rdc_status_t rdc_group_get_all_ids(
rdc_gpu_group_t group_id_list[], uint32_t* count) override;
rdc_status_t rdc_group_field_create(uint32_t num_field_ids,
uint32_t* field_ids, const char* field_group_name,
@@ -51,6 +53,8 @@ class RdcGroupSettingsImpl: public RdcGroupSettings {
rdc_status_t rdc_group_field_get_info(
rdc_field_grp_t rdc_field_group_id,
rdc_field_group_info_t* field_group_info) override;
rdc_status_t rdc_group_field_get_all_ids(
rdc_field_grp_t field_group_id_list[], uint32_t* count) override;
RdcGroupSettingsImpl();
projects/rdc/include/rdc_lib/impl/RdcStandaloneHandler.h
+11 -7
View File
@@ -40,9 +40,9 @@ class RdcStandaloneHandler: public RdcHandler {
rdc_status_t rdc_job_stop_stats(char job_id[64]) override;
// Discovery RdcAPI
rdc_status_t rdc_get_all_devices(
rdc_status_t rdc_device_get_all(
uint32_t gpu_index_list[RDC_MAX_NUM_DEVICES], uint32_t* count) override;
rdc_status_t rdc_get_device_attributes(uint32_t gpu_index,
rdc_status_t rdc_device_get_attributes(uint32_t gpu_index,
rdc_device_attributes_t* p_rdc_attr) override;
// Group RdcAPI
@@ -59,25 +59,29 @@ class RdcStandaloneHandler: public RdcHandler {
rdc_field_group_info_t* field_group_info) override;
rdc_status_t rdc_group_gpu_get_info(rdc_gpu_group_t p_rdc_group_id,
rdc_group_info_t* p_rdc_group_info) override;
rdc_status_t rdc_group_get_all_ids(
rdc_gpu_group_t group_id_list[], uint32_t* count) override;
rdc_status_t rdc_group_field_get_all_ids(
rdc_field_grp_t field_group_id_list[], uint32_t* count) override;
rdc_status_t rdc_group_gpu_destroy(
rdc_gpu_group_t p_rdc_group_id) override;
rdc_status_t rdc_group_field_destroy(
rdc_field_grp_t rdc_field_group_id) override;
// Field RdcAPI
rdc_status_t rdc_watch_fields(rdc_gpu_group_t group_id,
rdc_status_t rdc_field_watch(rdc_gpu_group_t group_id,
rdc_field_grp_t field_group_id, uint64_t update_freq,
double max_keep_age, uint32_t max_keep_samples) override;
rdc_status_t rdc_get_latest_value_for_field(uint32_t gpu_index,
rdc_status_t rdc_field_get_latest_value(uint32_t gpu_index,
uint32_t field, rdc_field_value* value) override;
rdc_status_t rdc_get_field_value_since(uint32_t gpu_index,
rdc_status_t rdc_field_get_value_since(uint32_t gpu_index,
uint32_t field, uint64_t since_time_stamp,
uint64_t *next_since_time_stamp, rdc_field_value* value) override;
rdc_status_t rdc_unwatch_fields(rdc_gpu_group_t group_id,
rdc_status_t rdc_field_unwatch(rdc_gpu_group_t group_id,
rdc_field_grp_t field_group_id) override;
// Control RdcAPI
rdc_status_t rdc_update_all_fields(uint32_t wait_for_update) override;
rdc_status_t rdc_field_update_all(uint32_t wait_for_update) override;
explicit RdcStandaloneHandler(const char* ip_and_port,
const char* root_ca, const char* client_cert, const char* client_key);
projects/rdc/include/rdc_lib/impl/RdcWatchTableImpl.h
+5 -5
View File
@@ -54,14 +54,14 @@ class RdcWatchTableImpl : public RdcWatchTable {
uint64_t update_freq, double max_keep_age,
uint32_t max_keep_samples) override;
rdc_status_t rdc_watch_fields(rdc_gpu_group_t group_id,
rdc_status_t rdc_field_watch(rdc_gpu_group_t group_id,
rdc_field_grp_t field_group_id, uint64_t update_freq,
double max_keep_age, uint32_t max_keep_samples) override;
//!< rdc_unwatch_fields() will not remove the entry from watch_table.
//!< rdc_field_unwatch() will not remove the entry from watch_table.
//!< The unwatched entry is still kept until the max_keep_age of the entry
//!< is reached, which will be handled in the clean_up() function.
rdc_status_t rdc_unwatch_fields(rdc_gpu_group_t group_id,
rdc_status_t rdc_field_unwatch(rdc_gpu_group_t group_id,
rdc_field_grp_t field_group_id) override;
//!< When the RDC is running as RDC_OPERATION_MODE_MANUAL, the user will
@@ -71,7 +71,7 @@ class RdcWatchTableImpl : public RdcWatchTable {
//!< This function may be called very frequently, and the cache cleanup
//!< is expensive. Internally, this function will throttle the cleanup to
//!< once per second.
rdc_status_t rdc_update_all_fields() override;
rdc_status_t rdc_field_update_all() override;
RdcWatchTableImpl(const RdcGroupSettingsPtr& group_settings,
const RdcCacheManagerPtr& cache_mgr,
@@ -97,7 +97,7 @@ class RdcWatchTableImpl : public RdcWatchTable {
std::map<RdcFieldKey, FieldSettings> watch_table_;
//!< The settings for each field can be deduced from watch_table. But every
//!< rdc_update_all_fields() call needs to deduce them. To improve the
//!< rdc_field_update_all() call needs to deduce them. To improve the
//!< performance, the fields_to_watch_ is used to track the field settings.
//!< Those settings will only be updated when watching or unwatching.
std::map<RdcFieldKey, FieldSettings> fields_to_watch_;