mirror of
https://github.com/wassname/ray.git
synced 2026-07-31 12:41:01 +08:00
Connect local scheduler to Plasma. (#11)
* Receive notifications about sealed objects from Plasma, and schedule tasks only when the dependencies are available locally. * Fix formatting. * Use version of Plasma with fix. * Fix. * Factor out the scheduling algorithm and use worker_index instead of the client socket to identify workers * Fixes * clang-format * fix remaining linter errors
This commit is contained in:
@@ -0,0 +1,5 @@
|
||||
BasedOnStyle: Chromium
|
||||
DerivePointerAlignment: false
|
||||
IndentCaseLabels: false
|
||||
PointerAlignment: Right
|
||||
SpaceAfterCStyleCast: true
|
||||
+11
@@ -45,6 +45,17 @@ matrix:
|
||||
|
||||
install:
|
||||
- ./install-dependencies.sh
|
||||
|
||||
# Install Plasma side by side.
|
||||
- cd ..
|
||||
- git clone https://github.com/ray-project/plasma.git
|
||||
- cd plasma
|
||||
- git checkout f189ca746b57f22371ef10077aa535492bbd8421
|
||||
- make
|
||||
- source setup-env.sh
|
||||
- cd ../photon
|
||||
|
||||
# Install Photon.
|
||||
- make
|
||||
- cd common/lib/python
|
||||
- python setup.py install --user
|
||||
|
||||
@@ -1,5 +1,5 @@
|
||||
CC = gcc
|
||||
CFLAGS = -g -Wall --std=c99 -D_XOPEN_SOURCE=500 -D_POSIX_C_SOURCE=200809L -Icommon/thirdparty -fPIC
|
||||
CFLAGS = -g -Wall --std=c99 -D_XOPEN_SOURCE=500 -D_POSIX_C_SOURCE=200809L -Icommon -Icommon/thirdparty -fPIC
|
||||
BUILD = build
|
||||
|
||||
all: $(BUILD)/photon_scheduler $(BUILD)/photon_client.a
|
||||
@@ -7,8 +7,8 @@ all: $(BUILD)/photon_scheduler $(BUILD)/photon_client.a
|
||||
$(BUILD)/photon_client.a: photon_client.o
|
||||
ar rcs $(BUILD)/photon_client.a photon_client.o
|
||||
|
||||
$(BUILD)/photon_scheduler: photon.h photon_scheduler.c common
|
||||
$(CC) $(CFLAGS) -o $@ photon_scheduler.c common/build/libcommon.a common/thirdparty/hiredis/libhiredis.a -Icommon/thirdparty -Icommon/
|
||||
$(BUILD)/photon_scheduler: photon.h photon_scheduler.c photon_algorithm.c common
|
||||
$(CC) $(CFLAGS) -o $@ photon_scheduler.c photon_algorithm.c common/build/libcommon.a common/thirdparty/hiredis/libhiredis.a -Icommon/thirdparty/ -Icommon/ ../plasma/build/libplasma_client.a -I../plasma/src/
|
||||
|
||||
common: FORCE
|
||||
git submodule update --init --recursive
|
||||
|
||||
+1
-1
Submodule common updated: 7be1a93d64...535bc8f0b8
@@ -1,6 +1,11 @@
|
||||
#ifndef PHOTON_H
|
||||
#define PHOTON_H
|
||||
|
||||
#include "common/task.h"
|
||||
#include "common/state/db.h"
|
||||
#include "utarray.h"
|
||||
#include "uthash.h"
|
||||
|
||||
enum photon_message_type {
|
||||
/** Notify the local scheduler that a task has finished. */
|
||||
TASK_DONE = 64,
|
||||
@@ -11,4 +16,25 @@ enum photon_message_type {
|
||||
EXECUTE_TASK,
|
||||
};
|
||||
|
||||
#endif
|
||||
// clang-format off
|
||||
/** Contains all information that is associated to a worker. */
|
||||
typedef struct {
|
||||
int sock;
|
||||
} worker;
|
||||
// clang-format on
|
||||
|
||||
/* These are needed to define the UT_arrays. */
|
||||
UT_icd task_ptr_icd;
|
||||
UT_icd worker_icd;
|
||||
|
||||
/** Resources that are exposed to the scheduling algorithm. */
|
||||
typedef struct {
|
||||
/** List of workers available to this node. The index into this array
|
||||
* is the worker_index and is used to identify workers throughout
|
||||
* the program. */
|
||||
UT_array *workers;
|
||||
/* The handle to the database. */
|
||||
db_handle *db;
|
||||
} scheduler_info;
|
||||
|
||||
#endif /* PHOTON_H */
|
||||
|
||||
@@ -0,0 +1,184 @@
|
||||
#include "photon_algorithm.h"
|
||||
|
||||
#include <stdbool.h>
|
||||
#include "utarray.h"
|
||||
|
||||
#include "state/task_log.h"
|
||||
#include "photon.h"
|
||||
#include "photon_scheduler.h"
|
||||
|
||||
typedef struct {
|
||||
/* Object id of this object. */
|
||||
object_id object_id;
|
||||
/* Handle for the uthash table. */
|
||||
UT_hash_handle handle;
|
||||
} available_object;
|
||||
|
||||
/** Part of the photon state that is maintained by the scheduling algorithm. */
|
||||
struct scheduler_state {
|
||||
/** An array of pointers to tasks that are waiting to be scheduled. */
|
||||
UT_array *task_queue;
|
||||
/** An array of worker indices corresponding to clients that are
|
||||
* waiting for tasks. */
|
||||
UT_array *available_workers;
|
||||
/** A hash map of the objects that are available in the local Plasma store.
|
||||
* This information could be a little stale. */
|
||||
available_object *local_objects;
|
||||
};
|
||||
|
||||
scheduler_state *make_scheduler_state(void) {
|
||||
scheduler_state *state = malloc(sizeof(scheduler_state));
|
||||
/* Initialize an empty hash map for the cache of local available objects. */
|
||||
state->local_objects = NULL;
|
||||
/* Initialize the local data structures used for queuing tasks and workers. */
|
||||
utarray_new(state->task_queue, &task_ptr_icd);
|
||||
utarray_new(state->available_workers, &ut_int_icd);
|
||||
return state;
|
||||
}
|
||||
|
||||
void free_scheduler_state(scheduler_state *s) {
|
||||
utarray_free(s->task_queue);
|
||||
utarray_free(s->available_workers);
|
||||
free(s);
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if all of the remote object arguments for a task are available in the
|
||||
* local object store.
|
||||
*
|
||||
* @param s The scheduler state.
|
||||
* @param task Task specification of the task to check.
|
||||
* @return This returns 1 if all of the remote object arguments for the task are
|
||||
* present in the local object store, otherwise it returns 0.
|
||||
*/
|
||||
bool can_run(scheduler_state *s, task_spec *task) {
|
||||
int64_t num_args = task_num_args(task);
|
||||
for (int i = 0; i < num_args; ++i) {
|
||||
if (task_arg_type(task, i) == ARG_BY_REF) {
|
||||
object_id obj_id = *task_arg_id(task, i);
|
||||
available_object *entry;
|
||||
HASH_FIND(handle, s->local_objects, &obj_id, sizeof(object_id), entry);
|
||||
if (entry == NULL) {
|
||||
/* The object is not present locally, so this task cannot be scheduled
|
||||
* right now. */
|
||||
return false;
|
||||
}
|
||||
}
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
/**
|
||||
* If there is a task whose dependencies are available locally, assign it to the
|
||||
* worker. This does not remove the worker from the available worker queue.
|
||||
*
|
||||
* @param s The scheduler state.
|
||||
* @param worker_index The index of the worker.
|
||||
* @return This returns 1 if it successfully assigned a task to the worker,
|
||||
* otherwise it returns 0.
|
||||
*/
|
||||
int find_and_schedule_task_if_possible(scheduler_info *info,
|
||||
scheduler_state *state,
|
||||
int worker_index) {
|
||||
int found_task_to_schedule = 0;
|
||||
/* Find the first task whose dependencies are available locally. */
|
||||
task_spec *spec;
|
||||
task_instance **task;
|
||||
int i = 0;
|
||||
for (; i < utarray_len(state->task_queue); ++i) {
|
||||
task = (task_instance **) utarray_eltptr(state->task_queue, i);
|
||||
spec = task_instance_task_spec(*task);
|
||||
if (can_run(state, spec)) {
|
||||
found_task_to_schedule = 1;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if (found_task_to_schedule) {
|
||||
/* This task's dependencies are available locally, so assign the task to the
|
||||
* worker. */
|
||||
assign_task_to_worker(info, spec, worker_index);
|
||||
/* Update the task queue data structure and free the task. */
|
||||
free(*task);
|
||||
utarray_erase(state->task_queue, i, 1);
|
||||
}
|
||||
return found_task_to_schedule;
|
||||
}
|
||||
|
||||
void handle_task_submitted(scheduler_info *info,
|
||||
scheduler_state *s,
|
||||
task_spec *task) {
|
||||
/* Create a unique task instance ID. This is different from the task ID and
|
||||
* is used to distinguish between potentially multiple executions of the
|
||||
* task. */
|
||||
task_iid task_iid = globally_unique_id();
|
||||
task_instance *instance =
|
||||
make_task_instance(task_iid, task, TASK_STATUS_WAITING, NIL_ID);
|
||||
/* If this task's dependencies are available locally, and if there is an
|
||||
* available worker, then assign this task to an available worker. Otherwise,
|
||||
* add this task to the local task queue. */
|
||||
int schedule_locally =
|
||||
(utarray_len(s->available_workers) > 0) && can_run(s, task);
|
||||
if (schedule_locally) {
|
||||
/* Get the last available worker in the available worker queue. */
|
||||
int *worker_index = (int *) utarray_back(s->available_workers);
|
||||
/* Tell the available worker to execute the task. */
|
||||
assign_task_to_worker(info, task, *worker_index);
|
||||
/* Remove the available worker from the queue and free the struct. */
|
||||
utarray_pop_back(s->available_workers);
|
||||
} else {
|
||||
/* Add the task to the task queue. This passes ownership of the task queue.
|
||||
* And the task will be freed when it is assigned to a worker. */
|
||||
utarray_push_back(s->task_queue, &instance);
|
||||
}
|
||||
/* Submit the task to redis. */
|
||||
task_log_add_task(info->db, instance);
|
||||
if (schedule_locally) {
|
||||
/* If the task was scheduled locally, we need to free it. Otherwise,
|
||||
* ownership of the task is passed to the task_queue, and it will be freed
|
||||
* when it is assigned to a worker. */
|
||||
free(instance);
|
||||
}
|
||||
}
|
||||
|
||||
void handle_worker_available(scheduler_info *info,
|
||||
scheduler_state *state,
|
||||
int worker_index) {
|
||||
int scheduled_task =
|
||||
find_and_schedule_task_if_possible(info, state, worker_index);
|
||||
/* If we couldn't find a task to schedule, add the worker to the queue of
|
||||
* available workers. */
|
||||
if (!scheduled_task) {
|
||||
for (int *p = (int *) utarray_front(state->available_workers); p != NULL;
|
||||
p = (int *) utarray_next(state->available_workers, p)) {
|
||||
CHECK(*p != worker_index);
|
||||
}
|
||||
/* Add client_sock to a list of available workers. This struct will be freed
|
||||
* when a task is assigned to this worker. */
|
||||
utarray_push_back(state->available_workers, &worker_index);
|
||||
LOG_INFO("Adding worker_index %d to available workers.\n", worker_index);
|
||||
}
|
||||
}
|
||||
|
||||
void handle_object_available(scheduler_info *info,
|
||||
scheduler_state *state,
|
||||
object_id object_id) {
|
||||
/* TODO(rkn): When does this get freed? */
|
||||
available_object *entry =
|
||||
(available_object *) malloc(sizeof(available_object));
|
||||
entry->object_id = object_id;
|
||||
HASH_ADD(handle, state->local_objects, object_id, sizeof(object_id), entry);
|
||||
|
||||
/* Check if we can schedule any tasks. */
|
||||
int num_tasks_scheduled = 0;
|
||||
for (int *p = (int *) utarray_front(state->available_workers); p != NULL;
|
||||
p = (int *) utarray_next(state->available_workers, p)) {
|
||||
/* Schedule a task on this worker if possible. */
|
||||
int scheduled_task = find_and_schedule_task_if_possible(info, state, *p);
|
||||
if (!scheduled_task) {
|
||||
/* There are no tasks we can schedule, so exit the loop. */
|
||||
break;
|
||||
}
|
||||
num_tasks_scheduled += 1;
|
||||
}
|
||||
utarray_erase(state->available_workers, 0, num_tasks_scheduled);
|
||||
}
|
||||
@@ -0,0 +1,88 @@
|
||||
#ifndef PHOTON_ALGORITHM_H
|
||||
#define PHOTON_ALGORITHM_H
|
||||
|
||||
#include "photon.h"
|
||||
#include "common/task.h"
|
||||
|
||||
/* ==== The scheduling algorithm ====
|
||||
*
|
||||
* This file contains declaration for all functions and data structures
|
||||
* that need to be provided if you want to implement a new algorithms
|
||||
* for the local scheduler.
|
||||
*
|
||||
*/
|
||||
|
||||
/** Internal state of the scheduling algorithm. */
|
||||
typedef struct scheduler_state scheduler_state;
|
||||
|
||||
/**
|
||||
* Initialize the scheduler state.
|
||||
*
|
||||
* @return Internal state of the scheduling algorithm.
|
||||
*/
|
||||
scheduler_state *make_scheduler_state(void);
|
||||
|
||||
/**
|
||||
* Free the scheduler state.
|
||||
*
|
||||
* @param state Internal state of the scheduling algorithm.
|
||||
* @return Void.
|
||||
*/
|
||||
void free_scheduler_state(scheduler_state *state);
|
||||
|
||||
/**
|
||||
* This function will be called when a new task is submitted by a worker for
|
||||
* execution.
|
||||
*
|
||||
* @param info Info about resources exposed by photon to the scheduling
|
||||
* algorithm.
|
||||
* @param state State of the scheduling algorithm.
|
||||
* @param task Task that is submitted by the worker.
|
||||
* @return Void.
|
||||
*/
|
||||
void handle_task_submitted(scheduler_info *info,
|
||||
scheduler_state *state,
|
||||
task_spec *task);
|
||||
|
||||
/**
|
||||
* This function will be called when a task is assigned by the global scheduler
|
||||
* for execution on this local scheduler.
|
||||
*
|
||||
* @param info Info about resources exposed by photon to the scheduling
|
||||
* algorithm.
|
||||
* @param state State of the scheduling algorithm.
|
||||
* @param task Task that is assigned by the global scheduler.
|
||||
* @return Void.
|
||||
*/
|
||||
void handle_task_assigned(scheduler_info *info,
|
||||
scheduler_state *state,
|
||||
task_spec *task);
|
||||
|
||||
/**
|
||||
* This function is called if a new object becomes available in the local
|
||||
* plasma store.
|
||||
*
|
||||
* @param info Info about resources exposed by photon to the scheduling
|
||||
* algorithm.
|
||||
* @param state State of the scheduling algorithm.
|
||||
* @param object_id ID of the object that became available.
|
||||
* @return Void.
|
||||
*/
|
||||
void handle_object_available(scheduler_info *info,
|
||||
scheduler_state *state,
|
||||
object_id object_id);
|
||||
|
||||
/**
|
||||
* This function is called when a new worker becomes available
|
||||
*
|
||||
* @param info Info about resources exposed by photon to the scheduling
|
||||
* algorithm.
|
||||
* @param state State of the scheduling algorithm.
|
||||
* @param worker_index The index of the worker that becomes available.
|
||||
* @return Void.
|
||||
*/
|
||||
void handle_worker_available(scheduler_info *info,
|
||||
scheduler_state *state,
|
||||
int worker_index);
|
||||
|
||||
#endif /* PHOTON_ALGORITHM_H */
|
||||
+99
-90
@@ -10,111 +10,92 @@
|
||||
#include "event_loop.h"
|
||||
#include "io.h"
|
||||
#include "photon.h"
|
||||
#include "photon_algorithm.h"
|
||||
#include "photon_scheduler.h"
|
||||
#include "plasma_client.h"
|
||||
#include "state/db.h"
|
||||
#include "state/task_log.h"
|
||||
#include "utarray.h"
|
||||
#include "uthash.h"
|
||||
|
||||
typedef struct {
|
||||
/** The file descriptor used to communicate with the worker. */
|
||||
int client_sock;
|
||||
} available_worker;
|
||||
|
||||
/* These are needed to define the UT_arrays. */
|
||||
UT_icd task_ptr_icd = {sizeof(task_instance *), NULL, NULL, NULL};
|
||||
UT_icd worker_icd = {sizeof(available_worker), NULL, NULL, NULL};
|
||||
UT_icd worker_icd = {sizeof(worker), NULL, NULL, NULL};
|
||||
|
||||
/** Association between the socket fd of a worker and its worker_index. */
|
||||
typedef struct {
|
||||
/** The socket fd of a worker. */
|
||||
int sock;
|
||||
/** The index of the worker in scheduler_info->workers. */
|
||||
int64_t worker_index;
|
||||
/** Handle for the hash table. */
|
||||
UT_hash_handle hh;
|
||||
} worker_index;
|
||||
|
||||
struct local_scheduler_state {
|
||||
/* The local scheduler event loop. */
|
||||
event_loop *loop;
|
||||
/* The handle to the database. */
|
||||
db_handle *db;
|
||||
/** This is an array of pointers to tasks that are waiting to be scheduled. */
|
||||
UT_array *task_queue;
|
||||
/** This is an array of file descriptors corresponding to clients that are
|
||||
* waiting for tasks. */
|
||||
UT_array *available_worker_queue;
|
||||
/* The Plasma client. */
|
||||
plasma_store_conn *plasma_conn;
|
||||
/* Association between client socket and worker index. */
|
||||
worker_index *worker_index;
|
||||
/* Info that is exposed to the scheduling algorithm. */
|
||||
scheduler_info *scheduler_info;
|
||||
/* State for the scheduling algorithm. */
|
||||
scheduler_state *scheduler_state;
|
||||
};
|
||||
|
||||
local_scheduler_state *
|
||||
init_local_scheduler(event_loop *loop, const char *redis_addr, int redis_port) {
|
||||
local_scheduler_state *init_local_scheduler(event_loop *loop,
|
||||
const char *redis_addr,
|
||||
int redis_port,
|
||||
const char *plasma_socket_name) {
|
||||
local_scheduler_state *state = malloc(sizeof(local_scheduler_state));
|
||||
state->loop = loop;
|
||||
state->db = db_connect(redis_addr, redis_port, "photon", "", -1);
|
||||
db_attach(state->db, loop);
|
||||
utarray_new(state->task_queue, &task_ptr_icd);
|
||||
utarray_new(state->available_worker_queue, &worker_icd);
|
||||
/* Connect to Plasma. This method will retry if Plasma hasn't started yet. */
|
||||
state->plasma_conn = plasma_store_connect(plasma_socket_name);
|
||||
/* Subscribe to notifications about sealed objects. */
|
||||
int plasma_fd = plasma_subscribe(state->plasma_conn);
|
||||
/* Add the callback that processes the notification to the event loop. */
|
||||
event_loop_add_file(loop, plasma_fd, EVENT_LOOP_READ,
|
||||
process_plasma_notification, state);
|
||||
state->worker_index = NULL;
|
||||
/* Add scheduler info. */
|
||||
state->scheduler_info = malloc(sizeof(scheduler_info));
|
||||
utarray_new(state->scheduler_info->workers, &worker_icd);
|
||||
/* Connect to Redis. */
|
||||
state->scheduler_info->db =
|
||||
db_connect(redis_addr, redis_port, "photon", "", -1);
|
||||
db_attach(state->scheduler_info->db, loop);
|
||||
/* Add scheduler state. */
|
||||
state->scheduler_state = make_scheduler_state();
|
||||
return state;
|
||||
};
|
||||
|
||||
void free_local_scheduler(local_scheduler_state *s) {
|
||||
db_disconnect(s->db);
|
||||
utarray_free(s->task_queue);
|
||||
utarray_free(s->available_worker_queue);
|
||||
db_disconnect(s->scheduler_info->db);
|
||||
free(s->scheduler_info);
|
||||
free_scheduler_state(s->scheduler_state);
|
||||
event_loop_destroy(s->loop);
|
||||
free(s);
|
||||
}
|
||||
|
||||
void handle_submit_task(local_scheduler_state *s, task_spec *task) {
|
||||
/* Create a unique task instance ID. This is different from the task ID and
|
||||
* is used to distinguish between potentially multiple executions of the
|
||||
* task. */
|
||||
task_iid task_iid = globally_unique_id();
|
||||
task_instance *instance =
|
||||
make_task_instance(task_iid, task, TASK_STATUS_WAITING, NIL_ID);
|
||||
/* Assign this task to an available worker. If there are no available workers,
|
||||
* then add this task to the local task queue. */
|
||||
int schedule_locally = utarray_len(s->available_worker_queue) > 0;
|
||||
if (schedule_locally) {
|
||||
/* Get the last available worker in the available worker queue. */
|
||||
available_worker *worker =
|
||||
(available_worker *)utarray_back(s->available_worker_queue);
|
||||
/* Tell the available worker to execute the task. */
|
||||
write_message(worker->client_sock, EXECUTE_TASK, task_size(task),
|
||||
(uint8_t *)task);
|
||||
/* Remove the available worker from the queue and free the struct. */
|
||||
utarray_pop_back(s->available_worker_queue);
|
||||
} else {
|
||||
/* Add the task to the task queue. This passes ownership of the task queue.
|
||||
* And the task will be freed when it is assigned to a worker. */
|
||||
utarray_push_back(s->task_queue, &instance);
|
||||
}
|
||||
/* Submit the task to redis. */
|
||||
task_log_add_task(s->db, instance);
|
||||
if (schedule_locally) {
|
||||
/* If the task was scheduled locally, we need to free it. Otherwise,
|
||||
* ownership of the task is passed to the task_queue, and it will be freed
|
||||
* when it is assigned to a worker. */
|
||||
free(instance);
|
||||
}
|
||||
void assign_task_to_worker(scheduler_info *info,
|
||||
task_spec *task,
|
||||
int worker_index) {
|
||||
CHECK(worker_index < utarray_len(info->workers));
|
||||
worker *w = (worker *) utarray_eltptr(info->workers, worker_index);
|
||||
write_message(w->sock, EXECUTE_TASK, task_size(task), (uint8_t *) task);
|
||||
}
|
||||
|
||||
void handle_get_task(local_scheduler_state *s, int client_sock) {
|
||||
/* If there is an available task, assign that task to this worker. Otherwise
|
||||
* add the worker to the queue of available workers. */
|
||||
if (utarray_len(s->task_queue) > 0) {
|
||||
/* Get the last task in the task queue. */
|
||||
task_instance **back = (task_instance **)utarray_back(s->task_queue);
|
||||
task_spec *task = task_instance_task_spec(*back);
|
||||
/* Send a task to the worker. */
|
||||
write_message(client_sock, EXECUTE_TASK, task_size(task), (uint8_t *)task);
|
||||
/* Update the task queue data structure and free the task. */
|
||||
utarray_pop_back(s->task_queue);
|
||||
free(*back);
|
||||
} else {
|
||||
/* Check that client_sock is not already in the available workers. */
|
||||
for (available_worker *p =
|
||||
(available_worker *)utarray_front(s->available_worker_queue);
|
||||
p != NULL;
|
||||
p = (available_worker *)utarray_next(s->available_worker_queue, p)) {
|
||||
CHECK(p->client_sock != client_sock);
|
||||
}
|
||||
/* Add client_sock to a list of available workers. This struct will be freed
|
||||
* when a task is assigned to this worker. */
|
||||
available_worker worker_info = {.client_sock = client_sock};
|
||||
utarray_push_back(s->available_worker_queue, &worker_info);
|
||||
LOG_INFO("Adding client_sock %d to available workers.\n", client_sock);
|
||||
}
|
||||
void process_plasma_notification(event_loop *loop,
|
||||
int client_sock,
|
||||
void *context,
|
||||
int events) {
|
||||
local_scheduler_state *s = context;
|
||||
/* Read the notification from Plasma. */
|
||||
uint8_t *message = (uint8_t *) malloc(sizeof(object_id));
|
||||
recv(client_sock, message, sizeof(object_id), 0);
|
||||
object_id *obj_id = (object_id *) message;
|
||||
handle_object_available(s->scheduler_info, s->scheduler_state, *obj_id);
|
||||
}
|
||||
|
||||
void process_message(event_loop *loop, int client_sock, void *context,
|
||||
@@ -126,16 +107,22 @@ void process_message(event_loop *loop, int client_sock, void *context,
|
||||
int64_t length;
|
||||
read_message(client_sock, &type, &length, &message);
|
||||
|
||||
LOG_DEBUG("New event of type %" PRId64, type);
|
||||
|
||||
switch (type) {
|
||||
case SUBMIT_TASK: {
|
||||
task_spec *task = (task_spec *)message;
|
||||
CHECK(task_size(task) == length);
|
||||
handle_submit_task(s, task);
|
||||
task_spec *spec = (task_spec *) message;
|
||||
CHECK(task_size(spec) == length);
|
||||
handle_task_submitted(s->scheduler_info, s->scheduler_state, spec);
|
||||
} break;
|
||||
case TASK_DONE: {
|
||||
} break;
|
||||
case GET_TASK: {
|
||||
handle_get_task(s, client_sock);
|
||||
worker_index *wi;
|
||||
HASH_FIND_INT(s->worker_index, &client_sock, wi);
|
||||
printf("worker_index is %" PRId64 "\n", wi->worker_index);
|
||||
handle_worker_available(s->scheduler_info, s->scheduler_state,
|
||||
wi->worker_index);
|
||||
} break;
|
||||
case DISCONNECT_CLIENT: {
|
||||
LOG_INFO("Disconnecting client on fd %d", client_sock);
|
||||
@@ -156,6 +143,14 @@ void new_client_connection(event_loop *loop, int listener_sock, void *context,
|
||||
int new_socket = accept_client(listener_sock);
|
||||
event_loop_add_file(loop, new_socket, EVENT_LOOP_READ, process_message, s);
|
||||
LOG_INFO("new connection with fd %d", new_socket);
|
||||
/* Add worker to list of workers. */
|
||||
/* TODO(pcm): Where shall we free this? */
|
||||
worker_index *new_worker_index = malloc(sizeof(worker_index));
|
||||
new_worker_index->sock = new_socket;
|
||||
new_worker_index->worker_index = utarray_len(s->scheduler_info->workers);
|
||||
HASH_ADD_INT(s->worker_index, sock, new_worker_index);
|
||||
worker worker = {.sock = new_socket};
|
||||
utarray_push_back(s->scheduler_info->workers, &worker);
|
||||
}
|
||||
|
||||
/* We need this code so we can clean up when we get a SIGTERM signal. */
|
||||
@@ -171,11 +166,14 @@ void signal_handler(int signal) {
|
||||
|
||||
/* End of the cleanup code. */
|
||||
|
||||
void start_server(const char *socket_name, const char *redis_addr,
|
||||
int redis_port) {
|
||||
void start_server(const char *socket_name,
|
||||
const char *redis_addr,
|
||||
int redis_port,
|
||||
const char *plasma_socket_name) {
|
||||
int fd = bind_ipc_sock(socket_name);
|
||||
event_loop *loop = event_loop_create();
|
||||
g_state = init_local_scheduler(loop, redis_addr, redis_port);
|
||||
g_state =
|
||||
init_local_scheduler(loop, redis_addr, redis_port, plasma_socket_name);
|
||||
|
||||
/* Run event loop. */
|
||||
event_loop_add_file(loop, fd, EVENT_LOOP_READ, new_client_connection,
|
||||
@@ -189,8 +187,10 @@ int main(int argc, char *argv[]) {
|
||||
char *scheduler_socket_name = NULL;
|
||||
/* IP address and port of redis. */
|
||||
char *redis_addr_port = NULL;
|
||||
/* Socket name for the local Plasma store. */
|
||||
char *plasma_socket_name = NULL;
|
||||
int c;
|
||||
while ((c = getopt(argc, argv, "s:r:")) != -1) {
|
||||
while ((c = getopt(argc, argv, "s:r:p:")) != -1) {
|
||||
switch (c) {
|
||||
case 's':
|
||||
scheduler_socket_name = optarg;
|
||||
@@ -198,6 +198,9 @@ int main(int argc, char *argv[]) {
|
||||
case 'r':
|
||||
redis_addr_port = optarg;
|
||||
break;
|
||||
case 'p':
|
||||
plasma_socket_name = optarg;
|
||||
break;
|
||||
default:
|
||||
LOG_ERR("unknown option %c", c);
|
||||
exit(-1);
|
||||
@@ -207,6 +210,11 @@ int main(int argc, char *argv[]) {
|
||||
LOG_ERR("please specify socket for incoming connections with -s switch");
|
||||
exit(-1);
|
||||
}
|
||||
if (!plasma_socket_name) {
|
||||
LOG_ERR("please specify socket for connecting to Plasma with -p switch");
|
||||
exit(-1);
|
||||
}
|
||||
/* Parse the Redis address into an IP address and a port. */
|
||||
char redis_addr[16] = {0};
|
||||
char redis_port[6] = {0};
|
||||
if (!redis_addr_port ||
|
||||
@@ -215,5 +223,6 @@ int main(int argc, char *argv[]) {
|
||||
LOG_ERR("need to specify redis address like 127.0.0.1:6379 with -r switch");
|
||||
exit(-1);
|
||||
}
|
||||
start_server(scheduler_socket_name, &redis_addr[0], atoi(redis_port));
|
||||
start_server(scheduler_socket_name, &redis_addr[0], atoi(redis_port),
|
||||
plasma_socket_name);
|
||||
}
|
||||
|
||||
+22
-9
@@ -2,6 +2,7 @@
|
||||
#define PHOTON_SCHEDULER_H
|
||||
|
||||
#include "task.h"
|
||||
#include "event_loop.h"
|
||||
|
||||
typedef struct local_scheduler_state local_scheduler_state;
|
||||
|
||||
@@ -15,25 +16,37 @@ typedef struct local_scheduler_state local_scheduler_state;
|
||||
* @param events Flag for events that are available on the listener socket.
|
||||
* @return Void.
|
||||
*/
|
||||
void new_client_connection(event_loop *loop, int listener_sock, void *context,
|
||||
void new_client_connection(event_loop *loop,
|
||||
int listener_sock,
|
||||
void *context,
|
||||
int events);
|
||||
|
||||
/**
|
||||
* Assign a task to a worker.
|
||||
* This function can be called by the scheduling algorithm to assign a task
|
||||
* to a worker.
|
||||
*
|
||||
* @param s State of the local scheduler.
|
||||
* @param client_sock Socket by which the worker is connected.
|
||||
* @param info
|
||||
* @param task The task that is submitted to the worker.
|
||||
* @param worker_index The index of the worker the task is submitted to.
|
||||
* @return Void.
|
||||
*/
|
||||
void handle_get_task(local_scheduler_state *s, int client_sock);
|
||||
void assign_task_to_worker(scheduler_info *info,
|
||||
task_spec *task,
|
||||
int worker_index);
|
||||
|
||||
/**
|
||||
* Handle incoming submit request by a worker.
|
||||
* This is the callback that is used to process a notification from the Plasma
|
||||
* store that an object has been sealed.
|
||||
*
|
||||
* @param s State of the local scheduler.
|
||||
* @param task Task specification of the task to be submitted.
|
||||
* @param loop The local scheduler's event loop.
|
||||
* @param client_sock The file descriptor to read the notification from.
|
||||
* @param context The local scheduler state.
|
||||
* @param events
|
||||
* @return Void.
|
||||
*/
|
||||
void handle_submit_task(local_scheduler_state *s, task_spec *task);
|
||||
void process_plasma_notification(event_loop *loop,
|
||||
int client_sock,
|
||||
void *context,
|
||||
int events);
|
||||
|
||||
#endif /* PHOTON_SCHEDULER_H */
|
||||
|
||||
+45
-8
@@ -6,9 +6,11 @@ import subprocess
|
||||
import sys
|
||||
import unittest
|
||||
import random
|
||||
import threading
|
||||
import time
|
||||
|
||||
import photon
|
||||
import plasma
|
||||
|
||||
USE_VALGRIND = False
|
||||
|
||||
@@ -18,14 +20,19 @@ class TestPhotonClient(unittest.TestCase):
|
||||
# Start Redis.
|
||||
redis_executable = os.path.join(os.path.abspath(os.path.dirname(__file__)), "../common/thirdparty/redis-3.2.3/src/redis-server")
|
||||
self.p1 = subprocess.Popen([redis_executable, "--loglevel", "warning"])
|
||||
# Start Plasma.
|
||||
plasma_executable = os.path.join(os.path.abspath(os.path.dirname(__file__)), "../../plasma/build/plasma_store")
|
||||
plasma_socket = "/tmp/plasma_store{}".format(random.randint(0, 10000))
|
||||
self.p2 = subprocess.Popen([plasma_executable, "-s", plasma_socket])
|
||||
time.sleep(0.1)
|
||||
self.plasma_client = plasma.PlasmaClient(plasma_socket)
|
||||
scheduler_executable = os.path.join(os.path.abspath(os.path.dirname(__file__)), "../build/photon_scheduler")
|
||||
scheduler_name = "/tmp/scheduler{}".format(random.randint(0, 10000))
|
||||
command = [scheduler_executable, "-s", scheduler_name, "-r", "127.0.0.1:6379"]
|
||||
command = [scheduler_executable, "-s", scheduler_name, "-r", "127.0.0.1:6379", "-p", plasma_socket]
|
||||
if USE_VALGRIND:
|
||||
self.p2 = subprocess.Popen(["valgrind", "--track-origins=yes", "--leak-check=full", "--show-leak-kinds=all"] + command)
|
||||
self.p3 = subprocess.Popen(["valgrind", "--track-origins=yes", "--leak-check=full", "--show-leak-kinds=all"] + command)
|
||||
else:
|
||||
self.p2 = subprocess.Popen(command)
|
||||
self.p3 = subprocess.Popen(command)
|
||||
if USE_VALGRIND:
|
||||
time.sleep(1.0)
|
||||
else:
|
||||
@@ -36,21 +43,30 @@ class TestPhotonClient(unittest.TestCase):
|
||||
def tearDown(self):
|
||||
# Kill the Redis server.
|
||||
self.p1.kill()
|
||||
# Kill Plasma.
|
||||
self.p2.kill()
|
||||
# Kill the local scheduler.
|
||||
if USE_VALGRIND:
|
||||
self.p2.send_signal(signal.SIGTERM)
|
||||
self.p2.wait()
|
||||
os._exit(self.p2.returncode)
|
||||
self.p3.send_signal(signal.SIGTERM)
|
||||
self.p3.wait()
|
||||
os._exit(self.p3.returncode)
|
||||
else:
|
||||
self.p2.kill()
|
||||
|
||||
self.p3.kill()
|
||||
|
||||
def test_submit_and_get_task(self):
|
||||
# TODO(rkn): This should be a FunctionID.
|
||||
function_id = photon.ObjectID(20 * "a")
|
||||
object_ids = [photon.ObjectID(20 * chr(i)) for i in range(256)]
|
||||
# Create and seal the objects in the object store so that we can schedule
|
||||
# all of the subsequent tasks.
|
||||
for object_id in object_ids:
|
||||
self.plasma_client.create(object_id.id(), 0)
|
||||
self.plasma_client.seal(object_id.id())
|
||||
# Define some arguments to use for the tasks.
|
||||
args_list = [
|
||||
[],
|
||||
#{},
|
||||
#(),
|
||||
1 * [1],
|
||||
10 * [1],
|
||||
100 * [1],
|
||||
@@ -104,6 +120,27 @@ class TestPhotonClient(unittest.TestCase):
|
||||
for num_return_vals in [0, 1, 2, 3, 5, 10, 100]:
|
||||
new_task = self.photon_client.get_task()
|
||||
|
||||
def test_scheduling_when_objects_ready(self):
|
||||
# Create a task and submit it.
|
||||
object_id = photon.ObjectID(20 * chr(0))
|
||||
# TODO(rkn): This should be a FunctionID.
|
||||
function_id = photon.ObjectID(20 * "a")
|
||||
task = photon.Task(function_id, [object_id], 0)
|
||||
self.photon_client.submit(task)
|
||||
# Launch a thread to get the task.
|
||||
def get_task():
|
||||
self.photon_client.get_task()
|
||||
t = threading.Thread(target=get_task)
|
||||
t.start()
|
||||
# Sleep to give the thread time to call get_task.
|
||||
time.sleep(0.1)
|
||||
# Create and seal the object ID in the object store. This should trigger a
|
||||
# scheduling event.
|
||||
self.plasma_client.create(object_id.id(), 0)
|
||||
self.plasma_client.seal(object_id.id())
|
||||
# Wait until the thread finishes so that we know the task was scheduled.
|
||||
t.join()
|
||||
|
||||
if __name__ == "__main__":
|
||||
if len(sys.argv) > 1:
|
||||
# pop the argument so we don't mess with unittest's own argument parser
|
||||
|
||||
Reference in New Issue
Block a user