Update DESIGN.txt and code accordingly
This commit is contained in:
+24
-8
@@ -170,14 +170,14 @@ Chunk Management:
|
||||
MS may add a subset of cs_add_list to ms_add_list based on the chunk replication and distribution policy
|
||||
MS sends ms_add_list and ms_rem_list to CS
|
||||
CS (1) Adds all elements from ms_rem_list to cs_rem_list
|
||||
(2) Adds elements in cs_add_list that are not in ms_add_list to cs_rem_list
|
||||
(3) Removes elements that are in ms_add_list and in cs_rem_list from cs_rem_list
|
||||
(4) Elements in ms_add_list that are not held by the chunk server are added to
|
||||
(2) Elements in ms_add_list that are not held by the chunk server are added to
|
||||
a temporary list tmp_list
|
||||
(5) Elements in cs_lst_list are added to tmp_list
|
||||
(3) Removes elements in ms_add_list from cs_add_list and cs_rem_list, then merges cs_add_list into cs_rem_list and clears cs_add_list.
|
||||
(4) Elements in cs_lst_list are added to tmp_list, then cs_lst_list is cleared.
|
||||
(6) tmp_list is sent to MS
|
||||
(7) cs_add_list is cleared
|
||||
MS (1) Receives tmp_list and sends download locations to CS for those chunks
|
||||
(2) Moves elements from ms_add_list that are not in tmp_list into ms_old_list
|
||||
(2) Merges ms_add_list into ms_old_list, then removes all items in tmp_list from ms_old_list
|
||||
(3) Sets ms_add_list equal to tmp_list
|
||||
|
||||
Chunk replication and distribution policy:
|
||||
@@ -190,6 +190,22 @@ Chunk Management:
|
||||
- The chunk is properly replicated with the new copy
|
||||
- The chunk is over-replicated with the new copy
|
||||
|
||||
If the chunk is useless, it is added to the cs_rem_list. If the chunk is
|
||||
over-replicated, it is added to the cs_rem_list of this chunk or any other
|
||||
holder of that chunk. Otherwise, add the chunk to the cs_add_list.
|
||||
If the chunk is not referenced by the file tree, do nothing.
|
||||
|
||||
If the chunk is properly replicated or under-replicated, add it to the ms_add_list.
|
||||
|
||||
If the chunk is over-replicated, either don't add it to the ms_add_list or add it to the ms_rem_list of some other holder.
|
||||
|
||||
TODO: The way chunk servers tell about the chunks they are holding
|
||||
recently changed. Now instead of sending a full chunk list when
|
||||
connecting, they send batches of chunks to the metadata server
|
||||
during state updates. It also changed that now chunk servers
|
||||
initiate updates.
|
||||
|
||||
When a chunk server connects (and authenticates)
|
||||
it sends alongside the auth message the hash of
|
||||
the hash list of all chunks. The metadata server
|
||||
then replies with a message saying whether it
|
||||
already cached that chunk list or not. If it didn't, the chunk server sends the entire list
|
||||
in chunks during state updates, with an increased
|
||||
update frequency.
|
||||
|
||||
@@ -9,6 +9,8 @@
|
||||
- add logging of:
|
||||
- when chunk servers connect and disconnect to the metadata server
|
||||
- Should list scenarios that need testing, like those where chunks would be dropped
|
||||
- Update DESIGN.txt and the code to remove the chunk list message. The information of chunks
|
||||
held by chunk servers is now transmitted to the metadata server during state updates
|
||||
|
||||
Roadmap:
|
||||
[ ] Complete all endpoints
|
||||
|
||||
+95
-193
@@ -321,139 +321,92 @@ static void start_download_if_necessary(ChunkServer *state)
|
||||
}
|
||||
}
|
||||
|
||||
static int
|
||||
process_metadata_server_state_update(ChunkServer *state, int conn_idx, ByteView msg)
|
||||
static int process_metadata_server_sync_2(ChunkServer *state,
|
||||
int conn_idx, ByteView msg)
|
||||
{
|
||||
BinaryReader reader = { msg.ptr, msg.len, 0 };
|
||||
|
||||
// Read header
|
||||
if (!binary_read(&reader, NULL, sizeof(MessageHeader)))
|
||||
return send_error(&state->tcp, conn_idx, true, MESSAGE_TYPE_STATE_UPDATE_ERROR, S("Invalid message"));
|
||||
return -1;
|
||||
|
||||
uint32_t add_count;
|
||||
if (!binary_read(&reader, &add_count, sizeof(add_count)))
|
||||
return send_error(&state->tcp, conn_idx, true, MESSAGE_TYPE_STATE_UPDATE_ERROR, S("Invalid message"));
|
||||
return -1;
|
||||
|
||||
HashList tmp_list;
|
||||
hash_list_init(&tmp_list);
|
||||
|
||||
for (uint32_t i = 0; i < add_count; i++) {
|
||||
|
||||
SHA256 hash;
|
||||
if (!binary_read(&reader, &hash, sizeof(hash)))
|
||||
return -1;
|
||||
|
||||
// Elements in ms_add_list that are not held by the
|
||||
// chunk server are added to a temporary list tmp_list
|
||||
|
||||
if (chunk_store_exists(&state->store, hash) < 0) { // TODO: does it return a boolean?
|
||||
if (hash_list_insert(&tmp_list, hash) < 0) {
|
||||
assert(0); // TODO
|
||||
}
|
||||
continue;
|
||||
}
|
||||
|
||||
hash_list_remove(&state->cs_rem_list, hash);
|
||||
hash_list_remove(&state->cs_add_list, hash);
|
||||
}
|
||||
|
||||
if (hash_list_merge(&state->cs_rem_list, state->cs_add_list) < 0) {
|
||||
assert(0); // TODO
|
||||
}
|
||||
|
||||
hash_list_clear(&state->cs_add_list);
|
||||
|
||||
uint32_t rem_count;
|
||||
if (!binary_read(&reader, &rem_count, sizeof(rem_count)))
|
||||
return send_error(&state->tcp, conn_idx, true, MESSAGE_TYPE_STATE_UPDATE_ERROR, S("Invalid message"));
|
||||
|
||||
SHA256 *add_list = sys_malloc(add_count * sizeof(SHA256));
|
||||
if (add_list == NULL)
|
||||
return send_error(&state->tcp, conn_idx, false, MESSAGE_TYPE_STATE_UPDATE_ERROR, S("Out of memory"));
|
||||
|
||||
SHA256 *rem_list = sys_malloc(rem_count * sizeof(SHA256));
|
||||
if (rem_list == NULL) {
|
||||
sys_free(add_list);
|
||||
return send_error(&state->tcp, conn_idx, false, MESSAGE_TYPE_STATE_UPDATE_ERROR, S("Out of memory"));
|
||||
}
|
||||
|
||||
for (uint32_t i = 0; i < add_count; i++) {
|
||||
if (!binary_read(&reader, &add_list[i], sizeof(SHA256))) {
|
||||
sys_free(add_list);
|
||||
sys_free(rem_list);
|
||||
return send_error(&state->tcp, conn_idx, true, MESSAGE_TYPE_STATE_UPDATE_ERROR, S("Invalid message"));
|
||||
}
|
||||
}
|
||||
return -1;
|
||||
|
||||
for (uint32_t i = 0; i < rem_count; i++) {
|
||||
if (!binary_read(&reader, &rem_list[i], sizeof(SHA256))) {
|
||||
sys_free(add_list);
|
||||
sys_free(rem_list);
|
||||
return send_error(&state->tcp, conn_idx, true, MESSAGE_TYPE_STATE_UPDATE_ERROR, S("Invalid message"));
|
||||
}
|
||||
}
|
||||
|
||||
if (binary_read(&reader, NULL, 1)) {
|
||||
sys_free(add_list);
|
||||
sys_free(rem_list);
|
||||
return send_error(&state->tcp, conn_idx, true, MESSAGE_TYPE_STATE_UPDATE_ERROR, S("Invalid message"));
|
||||
}
|
||||
|
||||
// Check that all items in the add_list are in the chunk directory
|
||||
// Any hashes that are missing are added to a missing list
|
||||
SHA256 *missing = NULL;
|
||||
uint32_t num_missing = 0;
|
||||
|
||||
Time current_time = get_current_time();
|
||||
|
||||
for (uint32_t i = 0; i < add_count; i++) {
|
||||
// If chunk is in removal list, unmark it (remove from removal list)
|
||||
removal_list_remove(&state->removal_list, add_list[i]);
|
||||
|
||||
// Check if chunk exists in the chunk store
|
||||
if (!chunk_store_exists(&state->store, add_list[i])) {
|
||||
|
||||
// Chunk is missing, add to missing list
|
||||
|
||||
if (missing == NULL) {
|
||||
missing = sys_malloc(add_count * sizeof(SHA256));
|
||||
if (missing == NULL) {
|
||||
assert(0); // TODO
|
||||
}
|
||||
}
|
||||
missing[num_missing++] = add_list[i];
|
||||
}
|
||||
}
|
||||
|
||||
// Append items from the rem_list to the removal list with timestamps
|
||||
for (uint32_t i = 0; i < rem_count; i++) {
|
||||
if (removal_list_add(&state->removal_list, rem_list[i], current_time) < 0) {
|
||||
sys_free(add_list);
|
||||
sys_free(rem_list);
|
||||
sys_free(missing);
|
||||
return send_error(&state->tcp, conn_idx, false, MESSAGE_TYPE_STATE_UPDATE_ERROR, S("Out of memory"));
|
||||
}
|
||||
}
|
||||
|
||||
sys_free(add_list);
|
||||
sys_free(rem_list);
|
||||
|
||||
// Respond to the metadata server
|
||||
if (num_missing == 0) {
|
||||
|
||||
// No missing chunks, send success
|
||||
|
||||
ByteQueue *output = tcp_output_buffer(&state->tcp, conn_idx);
|
||||
assert(output);
|
||||
|
||||
MessageWriter writer;
|
||||
message_writer_init(&writer, output, MESSAGE_TYPE_STATE_UPDATE_SUCCESS);
|
||||
if (!message_writer_free(&writer))
|
||||
SHA256 hash;
|
||||
if (!binary_read(&reader, &hash, sizeof(hash)))
|
||||
return -1;
|
||||
|
||||
} else {
|
||||
|
||||
// Some chunks are missing, send error with missing list
|
||||
|
||||
ByteQueue *output = tcp_output_buffer(&state->tcp, conn_idx);
|
||||
assert(output);
|
||||
|
||||
MessageWriter writer;
|
||||
message_writer_init(&writer, output, MESSAGE_TYPE_STATE_UPDATE_ERROR);
|
||||
|
||||
// Write error message
|
||||
string error_msg = S("Missing chunks");
|
||||
uint16_t error_len = (uint16_t)error_msg.len;
|
||||
message_write(&writer, &error_len, sizeof(error_len));
|
||||
message_write(&writer, error_msg.ptr, error_msg.len);
|
||||
|
||||
// Write missing count and missing hashes
|
||||
uint32_t tmp = num_missing;
|
||||
message_write(&writer, &tmp, sizeof(tmp));
|
||||
message_write(&writer, missing, num_missing * sizeof(SHA256));
|
||||
|
||||
sys_free(missing);
|
||||
|
||||
if (!message_writer_free(&writer))
|
||||
return -1;
|
||||
if (timed_hash_list_insert(&state->cs_rem_list, hash, current_time) < 0) {
|
||||
assert(0); // TODO
|
||||
}
|
||||
}
|
||||
|
||||
if (binary_read(&reader, NULL, 1))
|
||||
return -1;
|
||||
|
||||
ByteQueue *output = tcp_output_buffer(&state->tcp, conn_idx);
|
||||
assert(output);
|
||||
|
||||
MessageWriter writer;
|
||||
message_writer_init(&writer, output, MESSAGE_TYPE_SYNC_3);
|
||||
|
||||
uint32_t count = tmp_list.count + state->cs_lst_list.count; // TODO: overflow
|
||||
message_write(&writer, &count, sizeof(count));
|
||||
|
||||
for (uint32_t i = 0; i < tmp_list.count; i++) {
|
||||
SHA256 hash = tmp_list.items[i];
|
||||
message_write(&writer, &hash, sizeof(hash));
|
||||
}
|
||||
|
||||
for (uint32_t i = 0; i < state->cs_lst_list.count; i++) {
|
||||
SHA256 hash = state->cs_lst_list.items[i];
|
||||
message_write(&writer, &hash, sizeof(hash));
|
||||
}
|
||||
|
||||
if (!message_writer_free(&writer))
|
||||
return -1;
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int
|
||||
process_metadata_server_download_locations(ChunkServer *state, int conn_idx, ByteView msg)
|
||||
process_metadata_server_sync_4(ChunkServer *state, int conn_idx, ByteView msg)
|
||||
{
|
||||
(void) conn_idx;
|
||||
|
||||
@@ -565,92 +518,13 @@ process_metadata_server_download_locations(ChunkServer *state, int conn_idx, Byt
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int
|
||||
process_metadata_server_chunk_list_request(ChunkServer *state, int conn_idx, ByteView msg)
|
||||
{
|
||||
BinaryReader reader = { msg.ptr, msg.len, 0 };
|
||||
|
||||
// version
|
||||
if (!binary_read(&reader, NULL, sizeof(uint16_t)))
|
||||
return -1;
|
||||
|
||||
// type
|
||||
if (!binary_read(&reader, NULL, sizeof(uint16_t)))
|
||||
return -1;
|
||||
|
||||
// length
|
||||
if (!binary_read(&reader, NULL, sizeof(uint32_t)))
|
||||
return -1;
|
||||
|
||||
if (binary_read(&reader, NULL, 1))
|
||||
return 1;
|
||||
|
||||
ByteQueue *output = tcp_output_buffer(&state->tcp, conn_idx);
|
||||
assert(output);
|
||||
|
||||
MessageWriter writer;
|
||||
message_writer_init(&writer, output, MESSAGE_TYPE_CHUNK_LIST);
|
||||
|
||||
// Open the folder of chunks and write all hashes
|
||||
// to the metadata server. First, write the number
|
||||
// of hashes as a u32 integer, then that number
|
||||
// of hashes.
|
||||
// If the number is not known ahead of time, write
|
||||
// a dummy value and then patch it later.
|
||||
|
||||
ByteQueueOffset offset = byte_queue_offset(writer.output);
|
||||
|
||||
uint32_t num_hashes = 0; // Dummy value
|
||||
message_write(&writer, &num_hashes, sizeof(num_hashes));
|
||||
|
||||
DirectoryScanner scanner;
|
||||
string chunks_dir = (string) { state->store.path, strlen(state->store.path) };
|
||||
if (directory_scanner_init(&scanner, chunks_dir) < 0)
|
||||
return -1;
|
||||
|
||||
for (;;) {
|
||||
|
||||
string name;
|
||||
int ret = directory_scanner_next(&scanner, &name);
|
||||
|
||||
if (ret == 1)
|
||||
break;
|
||||
|
||||
if (ret < 0) {
|
||||
directory_scanner_free(&scanner);
|
||||
return -1;
|
||||
}
|
||||
|
||||
SHA256 hash;
|
||||
// TODO: file name to hash
|
||||
|
||||
message_write(&writer, &hash, sizeof(hash));
|
||||
num_hashes++;
|
||||
}
|
||||
directory_scanner_free(&scanner);
|
||||
|
||||
byte_queue_patch(writer.output, offset, &num_hashes, sizeof(num_hashes));
|
||||
|
||||
if (!message_writer_free(&writer))
|
||||
return -1;
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int
|
||||
process_metadata_server_message(ChunkServer *state, int conn_idx, uint16_t type, ByteView msg)
|
||||
{
|
||||
switch (type) {
|
||||
|
||||
case MESSAGE_TYPE_STATE_UPDATE:
|
||||
return process_metadata_server_state_update(state, conn_idx, msg);
|
||||
|
||||
case MESSAGE_TYPE_DOWNLOAD_LOCATIONS:
|
||||
return process_metadata_server_download_locations(state, conn_idx, msg);
|
||||
|
||||
case MESSAGE_TYPE_CHUNK_LIST_REQUEST:
|
||||
return process_metadata_server_chunk_list_request(state, conn_idx, msg);
|
||||
case MESSAGE_TYPE_SYNC_2: return process_metadata_server_sync_2(state, conn_idx, msg);
|
||||
case MESSAGE_TYPE_SYNC_4: return process_metadata_server_sync_4(state, conn_idx, msg);
|
||||
}
|
||||
|
||||
return -1;
|
||||
}
|
||||
|
||||
@@ -961,6 +835,26 @@ start_connecting_to_metadata_server(ChunkServer *state)
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int send_sync_message(ChunkServer *state)
|
||||
{
|
||||
assert(state->disconnect_time == INVALID_TIME);
|
||||
|
||||
MessageWriter writer;
|
||||
message_writer_init(&writer, output, MESSAGE_TYPE_SYNC);
|
||||
|
||||
uint32_t count = state->cs_add_list.count; // TODO: check implicit conversions
|
||||
message_write(&writer, &count, sizeof(count));
|
||||
|
||||
for (uint32_t i = 0; i < count; i++) {
|
||||
SHA256 hash = state->cs_add_list.items[i];
|
||||
message_write(&writer, &hash, sizeof(hash));
|
||||
}
|
||||
|
||||
if (!message_writer_free(&writer))
|
||||
return -1;
|
||||
return 0;
|
||||
}
|
||||
|
||||
int chunk_server_init(ChunkServer *state, int argc, char **argv, void **contexts, struct pollfd *polled, int *timeout)
|
||||
{
|
||||
string addr = getargs(argc, argv, "--addr", "127.0.0.1");
|
||||
@@ -1148,6 +1042,14 @@ int chunk_server_step(ChunkServer *state, void **contexts, struct pollfd *polled
|
||||
}
|
||||
}
|
||||
|
||||
// TODO: periodically send a SYNC message
|
||||
bool should_sync = false;
|
||||
if (should_sync) {
|
||||
if (send_sync_message(state) < 0) {
|
||||
assert(0); // TODO
|
||||
}
|
||||
}
|
||||
|
||||
// TODO: periodically look for chunks that have their hashes messed up and delete them
|
||||
|
||||
// Periodically retry pending downloads
|
||||
|
||||
+28
-10
@@ -3,6 +3,7 @@
|
||||
|
||||
#include <limits.h>
|
||||
|
||||
#include "metadata_server.h"
|
||||
#include "tcp.h"
|
||||
|
||||
#define TAG_METADATA_SERVER 1
|
||||
@@ -27,25 +28,42 @@ typedef struct {
|
||||
|
||||
typedef struct {
|
||||
SHA256 hash;
|
||||
Time marked_time;
|
||||
} PendingRemoval;
|
||||
Time time;
|
||||
} TimedHash;
|
||||
|
||||
typedef struct {
|
||||
int count;
|
||||
int capacity;
|
||||
PendingRemoval *items;
|
||||
} RemovalList;
|
||||
TimedHash *items;
|
||||
} TimedHashList;
|
||||
|
||||
typedef struct {
|
||||
bool trace;
|
||||
Address local_addr;
|
||||
Address remote_addr;
|
||||
Time disconnect_time;
|
||||
TCP tcp;
|
||||
|
||||
bool trace;
|
||||
|
||||
Address local_addr;
|
||||
|
||||
Address remote_addr;
|
||||
|
||||
Time disconnect_time;
|
||||
|
||||
TCP tcp;
|
||||
|
||||
ChunkStore store;
|
||||
|
||||
bool downloading;
|
||||
|
||||
PendingDownloadList pending_download_list;
|
||||
RemovalList removal_list;
|
||||
|
||||
// List of chunks added since the last update
|
||||
HashList cs_add_list;
|
||||
|
||||
// List of chunks marked for removal after a timeout
|
||||
TimedHashList cs_rem_list;
|
||||
|
||||
// List of chunks that were lost due to errors or forceful removals of chunk files
|
||||
HashList cs_lst_list;
|
||||
|
||||
} ChunkServer;
|
||||
|
||||
int chunk_server_init(ChunkServer *state, int argc, char **argv, void **contexts, struct pollfd *polled, int *timeout);
|
||||
|
||||
+3
-5
@@ -33,15 +33,13 @@ enum {
|
||||
MESSAGE_TYPE_WRITE_SUCCESS,
|
||||
|
||||
// Metadata server -> Chunk server
|
||||
MESSAGE_TYPE_STATE_UPDATE,
|
||||
MESSAGE_TYPE_SYNC_2,
|
||||
MESSAGE_TYPE_SYNC_4,
|
||||
MESSAGE_TYPE_DOWNLOAD_LOCATIONS,
|
||||
MESSAGE_TYPE_CHUNK_LIST_REQUEST,
|
||||
|
||||
// Chunk server -> Metadata server
|
||||
MESSAGE_TYPE_AUTH,
|
||||
MESSAGE_TYPE_CHUNK_LIST,
|
||||
MESSAGE_TYPE_STATE_UPDATE_ERROR,
|
||||
MESSAGE_TYPE_STATE_UPDATE_SUCCESS,
|
||||
MESSAGE_TYPE_SYNC,
|
||||
|
||||
// Chunk server -> Client
|
||||
MESSAGE_TYPE_CREATE_CHUNK_ERROR,
|
||||
|
||||
+120
-195
@@ -1,3 +1,8 @@
|
||||
#include "basic.h"
|
||||
#include "byte_queue.h"
|
||||
#include "config.h"
|
||||
#include "file_tree.h"
|
||||
#include "tcp.h"
|
||||
#define _GNU_SOURCE
|
||||
|
||||
#include <string.h>
|
||||
@@ -59,18 +64,18 @@ static void chunk_server_peer_init(ChunkServerPeer *chunk_server, Time current_t
|
||||
chunk_server->used = true;
|
||||
chunk_server->auth = false;
|
||||
chunk_server->num_addrs = 0;
|
||||
hash_list_init(&chunk_server->old_list);
|
||||
hash_list_init(&chunk_server->add_list);
|
||||
hash_list_init(&chunk_server->rem_list);
|
||||
hash_list_init(&chunk_server->ms_old_list);
|
||||
hash_list_init(&chunk_server->ms_add_list);
|
||||
hash_list_init(&chunk_server->ms_rem_list);
|
||||
chunk_server->last_sync_time = current_time;
|
||||
chunk_server->last_response_time = current_time;
|
||||
}
|
||||
|
||||
static void chunk_server_peer_free(ChunkServerPeer *chunk_server)
|
||||
{
|
||||
hash_list_free(&chunk_server->rem_list);
|
||||
hash_list_free(&chunk_server->add_list);
|
||||
hash_list_free(&chunk_server->old_list);
|
||||
hash_list_free(&chunk_server->ms_rem_list);
|
||||
hash_list_free(&chunk_server->ms_add_list);
|
||||
hash_list_free(&chunk_server->ms_old_list);
|
||||
chunk_server->used = false;
|
||||
}
|
||||
|
||||
@@ -284,6 +289,7 @@ process_client_delete(MetadataServer *state, int conn_idx, ByteView msg)
|
||||
if (binary_read(&reader, NULL, 1))
|
||||
return -1;
|
||||
|
||||
// TODO: return unused hashes and add them to the ms_rem_list of holder chunk servers
|
||||
int ret = file_tree_delete_entity(&state->file_tree, path);
|
||||
|
||||
if (ret < 0) {
|
||||
@@ -671,7 +677,7 @@ process_client_write(MetadataServer *state, int conn_idx, ByteView msg)
|
||||
int k = find_chunk_server_by_addr(state, results[i].addrs[j]);
|
||||
if (k == -1) return -1;
|
||||
|
||||
if (hash_list_insert(&state->chunk_servers[k].add_list, new_hashes[i]) < 0)
|
||||
if (hash_list_insert(&state->chunk_servers[k].ms_add_list, new_hashes[i]) < 0)
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
@@ -684,7 +690,7 @@ process_client_write(MetadataServer *state, int conn_idx, ByteView msg)
|
||||
// Add to rem_list for all chunk servers that have this chunk
|
||||
for (int j = 0; j < state->num_chunk_servers; j++) {
|
||||
if (chunk_server_peer_contains(&state->chunk_servers[j], removed_hash)) {
|
||||
if (!hash_list_insert(&state->chunk_servers[j].rem_list, removed_hash))
|
||||
if (!hash_list_insert(&state->chunk_servers[j].ms_rem_list, removed_hash))
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
@@ -726,36 +732,6 @@ chunk_server_from_conn(MetadataServer *state, int conn_idx)
|
||||
return &state->chunk_servers[tag];
|
||||
}
|
||||
|
||||
static int send_state_update(MetadataServer *state, int chunk_server_idx)
|
||||
{
|
||||
ChunkServerPeer *chunk_server = &state->chunk_servers[chunk_server_idx];
|
||||
|
||||
int conn_idx = tcp_index_from_tag(&state->tcp, chunk_server_idx);
|
||||
assert(conn_idx > -1);
|
||||
|
||||
ByteQueue *output = tcp_output_buffer(&state->tcp, conn_idx);
|
||||
assert(output);
|
||||
|
||||
MessageWriter writer;
|
||||
message_writer_init(&writer, output, MESSAGE_TYPE_STATE_UPDATE);
|
||||
|
||||
uint32_t add_count = chunk_server->add_list.count;
|
||||
uint32_t rem_count = chunk_server->rem_list.count;
|
||||
message_write(&writer, &add_count, sizeof(add_count));
|
||||
message_write(&writer, &rem_count, sizeof(rem_count));
|
||||
|
||||
for (uint32_t i = 0; i < add_count; i++)
|
||||
message_write(&writer, &chunk_server->add_list.items[i], sizeof(SHA256));
|
||||
|
||||
for (uint32_t i = 0; i < rem_count; i++)
|
||||
message_write(&writer, &chunk_server->rem_list.items[i], sizeof(SHA256));
|
||||
|
||||
if (!message_writer_free(&writer))
|
||||
return -1;
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int process_chunk_server_auth(MetadataServer *state,
|
||||
int conn_idx, ByteView msg)
|
||||
{
|
||||
@@ -838,172 +814,130 @@ static int process_chunk_server_auth(MetadataServer *state,
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int process_chunk_server_chunk_list(MetadataServer *state,
|
||||
static int process_chunk_server_sync(MetadataServer *state,
|
||||
int conn_idx, ByteView msg)
|
||||
{
|
||||
int chunk_server_idx = tcp_get_tag(&state->tcp, conn_idx);
|
||||
assert(chunk_server_idx > -1);
|
||||
assert(chunk_server_idx < MAX_CHUNK_SERVERS);
|
||||
assert(state->chunk_servers[chunk_server_idx].used);
|
||||
ChunkServerPeer *chunk_server = &state->chunk_servers[chunk_server_idx];
|
||||
assert(chunk_server_idx <= MAX_CHUNK_SERVERS);
|
||||
|
||||
// Iterate over each chunk held by this chunk
|
||||
// server and determine whether it's useless,
|
||||
// under-replicated, over-replicated, or
|
||||
// properly replicated.
|
||||
// Chunks that are useless or over-replicated
|
||||
// are added to the remove list of this chunk
|
||||
// server (note that any given chunk can only
|
||||
// be over-replicated by 1, so removing it from
|
||||
// this chunk server will fix the issue). If
|
||||
// the chunk is under-replicated or properly
|
||||
// replicated, do nothing. Note that it's only
|
||||
// possible for a chunk to be under-replicated
|
||||
// when the number of chunk servers is lower
|
||||
// than the replication factor.
|
||||
ChunkServerPeer *chunk_server = &state->chunk_servers[chunk_server_idx];
|
||||
assert(chunk_server->used);
|
||||
|
||||
BinaryReader reader = { msg.ptr, msg.len, 0 };
|
||||
|
||||
// version
|
||||
if (!binary_read(&reader, NULL, sizeof(uint16_t)))
|
||||
if (!binary_read(&reader, NULL, sizeof(MessageHeader)))
|
||||
return -1;
|
||||
|
||||
uint16_t type;
|
||||
if (!binary_read(&reader, &type, sizeof(type)))
|
||||
uint32_t count;
|
||||
if (!binary_read(&reader, &count, sizeof(count)))
|
||||
return -1;
|
||||
|
||||
if (type != MESSAGE_TYPE_CHUNK_LIST)
|
||||
return -1;
|
||||
|
||||
// length
|
||||
if (!binary_read(&reader, NULL, sizeof(uint32_t)))
|
||||
return -1;
|
||||
|
||||
uint32_t num_hashes;
|
||||
if (!binary_read(&reader, &num_hashes, sizeof(num_hashes)))
|
||||
return -1;
|
||||
|
||||
for (uint32_t i = 0; i < num_hashes; i++) {
|
||||
for (uint32_t i = 0; i < count; i++) {
|
||||
|
||||
SHA256 hash;
|
||||
if (!binary_read(&reader, &hash, sizeof(hash)))
|
||||
return -1;
|
||||
|
||||
bool drop = false;
|
||||
// If the chunk is not referenced by the file tree, do
|
||||
// nothing.
|
||||
|
||||
if (!file_tree_uses_hash(&state->file_tree, hash))
|
||||
drop = true;
|
||||
else {
|
||||
continue;
|
||||
|
||||
int holders[MAX_CHUNK_SERVERS];
|
||||
int num_holders = all_chunk_servers_holding_chunk(state, hash, holders, MAX_CHUNK_SERVERS);
|
||||
assert(num_holders > -1);
|
||||
assert(num_holders <= MAX_CHUNK_SERVERS);
|
||||
// If the chunk is properly replicated or under-replicated,
|
||||
// add it to the ms_add_list.
|
||||
|
||||
assert(num_holders <= state->replication_factor);
|
||||
int holders[MAX_CHUNK_SERVERS];
|
||||
int num_holders = all_chunk_servers_holding_chunk(state, hash, holders, MAX_CHUNK_SERVERS);
|
||||
assert(num_holders > -1);
|
||||
assert(num_holders <= MAX_CHUNK_SERVERS);
|
||||
|
||||
// Adding this chunk server as a holder will cause
|
||||
// the chunk to be over-replicated
|
||||
if (num_holders == state->replication_factor)
|
||||
drop = true;
|
||||
if (num_holders <= state->replication_factor) {
|
||||
if (hash_list_insert(&chunk_server->ms_add_list, hash) < 0) {
|
||||
assert(0); // TODO
|
||||
}
|
||||
continue;
|
||||
}
|
||||
|
||||
HashList *list;
|
||||
if (drop) list = &chunk_server->rem_list;
|
||||
else list = &chunk_server->add_list;
|
||||
// If the chunk is over-replicated, either don't add
|
||||
// it to the ms_add_list or add it to the ms_rem_list
|
||||
// of some other holder.
|
||||
|
||||
if (hash_list_insert(list, hash) < 0) {
|
||||
if (hash_list_insert(xxx, hash) < 0) {
|
||||
assert(0); // TODO
|
||||
}
|
||||
}
|
||||
|
||||
if (binary_read(&reader, NULL, 1))
|
||||
if (binary_read(&reader, NULL, 1)) // TODO: this should probably be an assertion
|
||||
return -1;
|
||||
|
||||
// No need to reply here
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int process_chunk_server_state_update_success(MetadataServer *state,
|
||||
int conn_idx, ByteView msg)
|
||||
{
|
||||
(void) msg; // Success message has no body
|
||||
ChunkServerPeer *chunk_server = chunk_server_from_conn(state, conn_idx);
|
||||
|
||||
// Merge add_list into old_list
|
||||
for (int i = 0; i < chunk_server->add_list.count; i++) {
|
||||
if (!hash_list_contains(&chunk_server->old_list, chunk_server->add_list.items[i]))
|
||||
hash_list_insert(&chunk_server->old_list, chunk_server->add_list.items[i]); // TODO: what if this fails?
|
||||
}
|
||||
|
||||
// Clear add_list and rem_list
|
||||
chunk_server->add_list.count = 0;
|
||||
chunk_server->rem_list.count = 0;
|
||||
|
||||
if (state->trace) {
|
||||
int tag = tcp_get_tag(&state->tcp, conn_idx);
|
||||
printf("Received STATE_UPDATE_SUCCESS from chunk server %d\n", tag);
|
||||
}
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int process_chunk_server_state_update_error(MetadataServer *state,
|
||||
int conn_idx, ByteView msg)
|
||||
{
|
||||
BinaryReader reader = { msg.ptr, msg.len, 0 };
|
||||
|
||||
// Read header
|
||||
if (!binary_read(&reader, NULL, sizeof(MessageHeader)))
|
||||
return -1;
|
||||
|
||||
// Read error message
|
||||
uint16_t error_len;
|
||||
if (!binary_read(&reader, &error_len, sizeof(error_len)))
|
||||
return -1;
|
||||
|
||||
char error_msg[256];
|
||||
int read_len = error_len < sizeof(error_msg) - 1 ? error_len : sizeof(error_msg) - 1;
|
||||
if (!binary_read(&reader, error_msg, read_len))
|
||||
return -1;
|
||||
error_msg[read_len] = '\0';
|
||||
|
||||
// Skip remaining error message if it was too long
|
||||
if (error_len > read_len)
|
||||
binary_read(&reader, NULL, error_len - read_len);
|
||||
|
||||
// Read missing chunks
|
||||
uint32_t num_missing;
|
||||
if (!binary_read(&reader, &num_missing, sizeof(num_missing)))
|
||||
return -1;
|
||||
|
||||
SHA256 *missing_chunks = sys_malloc(num_missing * sizeof(SHA256));
|
||||
if (missing_chunks == NULL)
|
||||
return -1;
|
||||
|
||||
for (uint32_t i = 0; i < num_missing; i++) {
|
||||
if (!binary_read(&reader, &missing_chunks[i], sizeof(SHA256))) {
|
||||
sys_free(missing_chunks);
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
|
||||
if (state->trace) {
|
||||
int tag = tcp_get_tag(&state->tcp, conn_idx);
|
||||
printf("Received STATE_UPDATE_ERROR from chunk server %d: %s (missing %u chunks)\n",
|
||||
tag, error_msg, num_missing);
|
||||
}
|
||||
// Respond with ms_add_list and ms_rem_list
|
||||
|
||||
ByteQueue *output = tcp_output_buffer(&state->tcp, conn_idx);
|
||||
assert(output);
|
||||
|
||||
MessageWriter writer;
|
||||
message_writer_init(&writer, output, MESSAGE_TYPE_DOWNLOAD_LOCATIONS);
|
||||
message_writer_init(&writer, output, MESSAGE_TYPE_SYNC_2);
|
||||
|
||||
message_write(&writer, &num_missing, sizeof(num_missing));
|
||||
uint32_t add_count = chunk_server->ms_add_list.count; // TODO: check implicit casts
|
||||
message_write(&writer, &add_count, sizeof(add_count));
|
||||
|
||||
for (uint32_t i = 0; i < num_missing; i++) {
|
||||
for (uint32_t i = 0; i < add_count; i++) {
|
||||
SHA256 hash = chunk_server->ms_add_list.items[i];
|
||||
message_write(&writer, &hash, sizeof(hash));
|
||||
}
|
||||
|
||||
SHA256 hash = missing_chunks[i];
|
||||
uint32_t rem_count = chunk_server->ms_rem_list.count; // TODO: check implicit casts
|
||||
message_write(&writer, &rem_count, sizeof(rem_count));
|
||||
|
||||
for (uint32_t i = 0; i < rem_count; i++) {
|
||||
SHA256 hash = chunk_server->ms_rem_list.items[i];
|
||||
message_write(&writer, &hash, sizeof(hash));
|
||||
}
|
||||
|
||||
if (!message_writer_free(&writer))
|
||||
return -1;
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int process_chunk_server_sync_3(MetadataServer *state,
|
||||
int conn_idx, ByteView msg)
|
||||
{
|
||||
int chunk_server_idx = tcp_get_tag(&state->tcp, conn_idx);
|
||||
assert(chunk_server_idx > -1);
|
||||
assert(chunk_server_idx <= MAX_CHUNK_SERVERS);
|
||||
|
||||
ChunkServerPeer *chunk_server = &state->chunk_servers[chunk_server_idx];
|
||||
assert(chunk_server->used);
|
||||
|
||||
BinaryReader reader = { msg.ptr, msg.len, 0 };
|
||||
|
||||
if (!binary_read(&reader, NULL, sizeof(MessageHeader)))
|
||||
return -1;
|
||||
|
||||
uint32_t count;
|
||||
if (!binary_read(&reader, &count, sizeof(count)))
|
||||
return -1;
|
||||
|
||||
ByteQueue *output = tcp_output_buffer(&state->tcp, conn_idx);
|
||||
assert(output);
|
||||
|
||||
MessageWriter writer;
|
||||
message_writer_init(&writer, output, MESSAGE_TYPE_SYNC_4);
|
||||
|
||||
HashList tmp_list;
|
||||
hash_list_init(&tmp_list);
|
||||
|
||||
for (uint32_t i = 0; i < count; i++) {
|
||||
|
||||
SHA256 hash;
|
||||
if (!binary_read(&reader, &hash, sizeof(hash)))
|
||||
return -1;
|
||||
|
||||
if (hash_list_insert(&tmp_list, hash) < 0) {
|
||||
assert(0); // TODO
|
||||
}
|
||||
|
||||
int holders[MAX_CHUNK_SERVERS];
|
||||
int num_holders = all_chunk_servers_holding_chunk(state, hash, holders, MAX_CHUNK_SERVERS);
|
||||
@@ -1013,17 +947,26 @@ static int process_chunk_server_state_update_error(MetadataServer *state,
|
||||
uint32_t tmp = num_holders;
|
||||
message_write(&writer, &tmp, sizeof(tmp));
|
||||
|
||||
for (int j = 0; j < num_holders; j++)
|
||||
message_write_server_addr(&writer, &state->chunk_servers[holders[j]]);
|
||||
|
||||
message_write(&writer, &hash, sizeof(hash));
|
||||
for (int i = 0; i < num_holders; i++)
|
||||
message_write_server_addr(&writer, xxx);
|
||||
}
|
||||
|
||||
sys_free(missing_chunks);
|
||||
if (binary_read(&reader, NULL, 1)) // TODO: this should probably be an assertion
|
||||
return -1;
|
||||
|
||||
if (hash_list_merge(&state->ms_old_list, state->ms_add_list) < 0) {
|
||||
assert(0); // TODO
|
||||
}
|
||||
|
||||
if (hash_list_remove_set(&state->ms_old_list, tmp_list) < 0) {
|
||||
assert(0); // TODO
|
||||
}
|
||||
|
||||
hash_list_free(&state->ms_add_list);
|
||||
state->ms_add_list = tmp_list;
|
||||
|
||||
if (!message_writer_free(&writer))
|
||||
return -1;
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
@@ -1035,14 +978,11 @@ process_chunk_server_message(MetadataServer *state,
|
||||
case MESSAGE_TYPE_AUTH:
|
||||
return process_chunk_server_auth(state, conn_idx, msg);
|
||||
|
||||
case MESSAGE_TYPE_CHUNK_LIST:
|
||||
return process_chunk_server_chunk_list(state, conn_idx, msg);
|
||||
case MESSAGE_TYPE_SYNC:
|
||||
return process_chunk_server_sync(state, conn_idx, msg);
|
||||
|
||||
case MESSAGE_TYPE_STATE_UPDATE_SUCCESS:
|
||||
return process_chunk_server_state_update_success(state, conn_idx, msg);
|
||||
|
||||
case MESSAGE_TYPE_STATE_UPDATE_ERROR:
|
||||
return process_chunk_server_state_update_error(state, conn_idx, msg);
|
||||
case MESSAGE_TYPE_SYNC_3:
|
||||
return process_chunk_server_sync_3(state, conn_idx, msg);
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
@@ -1051,8 +991,7 @@ static bool is_chunk_server_message_type(uint16_t type)
|
||||
{
|
||||
switch (type) {
|
||||
case MESSAGE_TYPE_AUTH:
|
||||
case MESSAGE_TYPE_STATE_UPDATE_ERROR:
|
||||
case MESSAGE_TYPE_STATE_UPDATE_SUCCESS:
|
||||
case MESSAGE_TYPE_SYNC:
|
||||
return true;
|
||||
|
||||
default:
|
||||
@@ -1115,7 +1054,6 @@ int metadata_server_step(MetadataServer *state, void **contexts, struct pollfd *
|
||||
Event events[MAX_CONNS+1];
|
||||
int num_events = tcp_translate_events(&state->tcp, events, contexts, polled, num_polled);
|
||||
|
||||
// Implement periodic health checks: send STATE_UPDATE to chunk servers
|
||||
Time current_time = get_current_time();
|
||||
if (current_time == INVALID_TIME) {
|
||||
assert(0); // TODO
|
||||
@@ -1213,19 +1151,6 @@ int metadata_server_step(MetadataServer *state, void **contexts, struct pollfd *
|
||||
continue;
|
||||
}
|
||||
nearest_deadline(&next_wakeup, response_timeout);
|
||||
|
||||
if (chunk_server->auth && chunk_server->last_sync_done) {
|
||||
Time sync_timeout = chunk_server->last_sync_time + (Time) SYNC_INTERVAL * 1000000000;
|
||||
if (current_time > sync_timeout) {
|
||||
if (send_state_update(state, i) < 0) {
|
||||
assert(0); // TODO
|
||||
}
|
||||
chunk_server->last_sync_time = current_time;
|
||||
chunk_server->last_sync_done = false;
|
||||
continue;
|
||||
}
|
||||
nearest_deadline(&next_wakeup, sync_timeout);
|
||||
}
|
||||
}
|
||||
|
||||
*timeout = deadline_to_timeout(next_wakeup, current_time);
|
||||
|
||||
@@ -23,17 +23,14 @@ typedef struct {
|
||||
int num_addrs;
|
||||
Address addrs[MAX_SERVER_ADDRS];
|
||||
|
||||
// Chunks held by the chunk server during
|
||||
// the last update
|
||||
HashList old_list;
|
||||
// List of chunks that are known to be held by CS
|
||||
HashList ms_old_list;
|
||||
|
||||
// Chunks added to the chunk server since
|
||||
// the last update
|
||||
HashList add_list;
|
||||
// List of chunks that should be held by CS
|
||||
HashList ms_add_list;
|
||||
|
||||
// Chunks removed from the chunk server
|
||||
// since the last update
|
||||
HashList rem_list;
|
||||
// List of chunks that may be held by CS but should removed from it
|
||||
HashList ms_rem_list;
|
||||
|
||||
// Time when last STATE_UPDATE was sent
|
||||
Time last_sync_time;
|
||||
|
||||
Reference in New Issue
Block a user