static void sparse_file_write_block(struct output_file *out,
struct backed_block *bb)
{
switch (backed_block_type(bb)) {
case BACKED_BLOCK_DATA:
write_data_chunk(out, backed_block_len(bb), backed_block_data(bb));
break;
case BACKED_BLOCK_FILE:
write_file_chunk(out, backed_block_len(bb),
backed_block_filename(bb), backed_block_file_offset(bb));
break;
case BACKED_BLOCK_FD:
write_fd_chunk(out, backed_block_len(bb),
backed_block_fd(bb), backed_block_file_offset(bb));
break;
case BACKED_BLOCK_FILL:
write_fill_chunk(out, backed_block_len(bb),
backed_block_fill_val(bb));
break;
}
}
STATIC mp_uint_t _webrepl_read(mp_obj_t self_in, void *buf, mp_uint_t size, int *errcode) {
// We know that os.dupterm always calls with size = 1
assert(size == 1);
mp_obj_webrepl_t *self = self_in;
const mp_stream_p_t *sock_stream = mp_get_stream(self->sock);
mp_uint_t out_sz = sock_stream->read(self->sock, buf, size, errcode);
//DEBUG_printf("webrepl: Read %d initial bytes from websocket\n", out_sz);
if (out_sz == 0 || out_sz == MP_STREAM_ERROR) {
return out_sz;
}
if (self->state == STATE_PASSWD) {
char c = *(char*)buf;
if (c == '\r' || c == '\n') {
self->hdr.fname[self->data_to_recv] = 0;
DEBUG_printf("webrepl: entered password: %s\n", self->hdr.fname);
if (strcmp(self->hdr.fname, webrepl_passwd) != 0) {
write_webrepl_str(self->sock, SSTR(denied_prompt));
return 0;
}
self->state = STATE_NORMAL;
self->data_to_recv = 0;
write_webrepl_str(self->sock, SSTR(connected_prompt));
} else if (self->data_to_recv < 10) {
self->hdr.fname[self->data_to_recv++] = c;
}
return -2;
}
// If last read data belonged to text record (== REPL)
int err;
if (sock_stream->ioctl(self->sock, MP_STREAM_GET_DATA_OPTS, 0, &err) == 1) {
return out_sz;
}
DEBUG_printf("webrepl: received bin data, hdr_to_recv: %d, data_to_recv=%d\n", self->hdr_to_recv, self->data_to_recv);
if (self->hdr_to_recv != 0) {
char *p = (char*)&self->hdr + sizeof(self->hdr) - self->hdr_to_recv;
*p++ = *(char*)buf;
if (--self->hdr_to_recv != 0) {
mp_uint_t hdr_sz = sock_stream->read(self->sock, p, self->hdr_to_recv, errcode);
if (hdr_sz == MP_STREAM_ERROR) {
return hdr_sz;
}
self->hdr_to_recv -= hdr_sz;
if (self->hdr_to_recv != 0) {
return -2;
}
}
DEBUG_printf("webrepl: op: %d, file: %s, chunk @%x, sz=%d\n", self->hdr.type, self->hdr.fname, (uint32_t)self->hdr.offset, self->hdr.size);
handle_op(self);
return -2;
}
if (self->data_to_recv != 0) {
static byte filebuf[512];
filebuf[0] = *(byte*)buf;
mp_uint_t buf_sz = 1;
if (--self->data_to_recv != 0) {
size_t to_read = MIN(sizeof(filebuf) - 1, self->data_to_recv);
mp_uint_t sz = sock_stream->read(self->sock, filebuf + 1, to_read, errcode);
if (sz == MP_STREAM_ERROR) {
return sz;
}
self->data_to_recv -= sz;
buf_sz += sz;
}
if (self->hdr.type == PUT_FILE) {
DEBUG_printf("webrepl: Writing %lu bytes to file\n", buf_sz);
int err;
mp_uint_t res = mp_stream_write_exactly(self->cur_file, filebuf, buf_sz, &err);
if (err != 0 || res != buf_sz) {
assert(0);
}
} else if (self->hdr.type == GET_FILE) {
assert(buf_sz == 1);
assert(self->data_to_recv == 0);
assert(filebuf[0] == 0);
mp_uint_t out_sz = write_file_chunk(self);
if (out_sz != 0) {
self->data_to_recv = 1;
}
}
if (self->data_to_recv == 0) {
mp_stream_close(self->cur_file);
self->hdr_to_recv = sizeof(struct webrepl_file);
DEBUG_printf("webrepl: Finished file operation %d\n", self->hdr.type);
write_webrepl_resp(self->sock, 0);
}
#ifdef MICROPY_PY_WEBREPL_DELAY
// Some platforms may have broken drivers and easily gets
// overloaded with modest traffic WebREPL file transfers
// generate. The basic workaround is a crude rate control
// done in such way.
mp_hal_delay_ms(MICROPY_PY_WEBREPL_DELAY);
#endif
}
return -2;
}
int main(int argc, char* argv[])
{
MPI_Init(&argc, &argv);
mfu_init();
/* get our rank and number of ranks in the job */
int rank, ranks;
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &ranks);
/* pointer to mfu_walk_opts */
mfu_walk_opts_t* walk_opts = mfu_walk_opts_new();
uint64_t idx;
int option_index = 0;
int usage = 0;
int report = 0;
unsigned int numpaths = 0;
mfu_param_path* paths = NULL;
unsigned long long bytes;
/* verbose by default */
mfu_debug_level = MFU_LOG_VERBOSE;
/* default to 1MB stripe size, stripe across all OSTs, and all files are candidates */
int stripes = -1;
uint64_t stripe_size = 1048576;
uint64_t min_size = 0;
static struct option long_options[] = {
{"count", 1, 0, 'c'},
{"size", 1, 0, 's'},
{"minsize", 1, 0, 'm'},
{"report", 0, 0, 'r'},
{"progress", 1, 0, 'P'},
{"verbose", 0, 0, 'v'},
{"quiet", 0, 0, 'q'},
{"help", 0, 0, 'h'},
{0, 0, 0, 0}
};
while (1) {
int c = getopt_long(argc, argv, "c:s:m:rvqh",
long_options, &option_index);
if (c == -1) {
break;
}
switch (c) {
case 'c':
/* stripe count */
stripes = atoi(optarg);
break;
case 's':
/* stripe size in bytes */
if (mfu_abtoull(optarg, &bytes) != MFU_SUCCESS) {
if (rank == 0) {
MFU_LOG(MFU_LOG_ERR, "Failed to parse stripe size: %s", optarg);
}
MPI_Abort(MPI_COMM_WORLD, 1);
}
stripe_size = (uint64_t)bytes;
break;
case 'm':
/* min file size in bytes */
if (mfu_abtoull(optarg, &bytes) != MFU_SUCCESS) {
if (rank == 0) {
MFU_LOG(MFU_LOG_ERR, "Failed to parse minimum file size: %s", optarg);
}
MPI_Abort(MPI_COMM_WORLD, 1);
}
min_size = (uint64_t)bytes;
break;
case 'r':
/* report striping info */
report = 1;
break;
case 'P':
mfu_progress_timeout = atoi(optarg);
break;
case 'v':
mfu_debug_level = MFU_LOG_VERBOSE;
break;
case 'q':
mfu_debug_level = MFU_LOG_NONE;
break;
case 'h':
/* display usage */
usage = 1;
break;
case '?':
/* display usage */
usage = 1;
break;
default:
if (rank == 0) {
printf("?? getopt returned character code 0%o ??\n", c);
}
}
}
/* check that we got a valid progress value */
if (mfu_progress_timeout < 0) {
if (rank == 0) {
MFU_LOG(MFU_LOG_ERR, "Seconds in --progress must be non-negative: %d invalid", mfu_progress_timeout);
}
usage = 1;
}
/* paths to walk come after the options */
if (optind < argc) {
/* determine number of paths specified by user */
numpaths = argc - optind;
/* allocate space for each path */
paths = (mfu_param_path*) MFU_MALLOC((size_t)numpaths * sizeof(mfu_param_path));
/* process each path */
char** p = &argv[optind];
mfu_param_path_set_all((uint64_t)numpaths, (const char**)p, paths);
optind += numpaths;
} else {
usage = 1;
}
/* if we need to print usage, print it and exit */
if (usage) {
if (rank == 0) {
print_usage();
}
mfu_finalize();
MPI_Finalize();
return 1;
}
/* nothing to do if lustre support is disabled */
#ifndef LUSTRE_SUPPORT
if (rank == 0) {
MFU_LOG(MFU_LOG_ERR, "Lustre support is disabled.");
}
MPI_Abort(MPI_COMM_WORLD, 1);
#endif
/* stripe count must be -1 for all available or greater than 0 */
if (stripes < -1) {
if (rank == 0) {
MFU_LOG(MFU_LOG_ERR, "Stripe count must be -1 for all servers, 0 for lustre file system default, or a positive value");
}
MPI_Abort(MPI_COMM_WORLD, 1);
}
/* lustre requires stripe sizes to be aligned */
if (stripe_size > 0 && stripe_size % 65536 != 0) {
if (rank == 0) {
MFU_LOG(MFU_LOG_ERR, "Stripe size must be a multiple of 65536");
}
MPI_Abort(MPI_COMM_WORLD, 1);
}
/* TODO: verify that source / target are on Lustre */
/* walk list of input paths and stat as we walk */
mfu_flist flist = mfu_flist_new();
mfu_flist_walk_param_paths(numpaths, paths, walk_opts, flist);
/* filter down our list to files which don't meet our striping requirements */
mfu_flist filtered = filter_list(flist, stripes, stripe_size, min_size, &create_prog_count_total, &stripe_prog_bytes_total);
mfu_flist_free(&flist);
MPI_Barrier(MPI_COMM_WORLD);
/* report the file size and stripe count of all files we found */
if (report) {
/* report the files in our filtered list */
stripe_info_report(filtered);
/* free the paths and our list */
mfu_flist_free(&filtered);
mfu_param_path_free_all(numpaths, paths);
mfu_free(&paths);
/* finalize */
mfu_finalize();
MPI_Finalize();
return 0;
}
/* generate a global suffix for our temp files and have each node check it's list */
char suffix[8];
uint64_t retry;
/* seed our random number generator */
srand(time(NULL));
/* keep trying to make a valid random suffix...*/
do {
uint64_t attempt = 0;
/* make rank 0 responsible for generating a random suffix */
if (rank == 0) {
generate_suffix(suffix, sizeof(suffix));
}
/* broadcast the random suffix to all ranks */
MPI_Bcast(suffix, sizeof(suffix), MPI_CHAR, 0, MPI_COMM_WORLD);
/* check that the file doesn't already exist */
uint64_t size = mfu_flist_size(filtered);
for (idx = 0; idx < size; idx++) {
char temp_path[PATH_MAX];
strcpy(temp_path, mfu_flist_file_get_name(filtered, idx));
strcat(temp_path, suffix);
if(!mfu_access(temp_path, F_OK)) {
/* the file already exists */
attempt = 1;
break;
}
}
/* do a reduce to figure out if a rank has a file collision */
MPI_Allreduce(&attempt, &retry, 1, MPI_UINT64_T, MPI_MAX, MPI_COMM_WORLD);
} while(retry != 0);
/* initialize progress messages while creating files */
create_prog_count = 0;
create_prog = mfu_progress_start(mfu_progress_timeout, 1, MPI_COMM_WORLD, create_progress_fn);
/* create new files so we can restripe */
uint64_t size = mfu_flist_size(filtered);
for (idx = 0; idx < size; idx++) {
char temp_path[PATH_MAX];
strcpy(temp_path, mfu_flist_file_get_name(filtered, idx));
strcat(temp_path, suffix);
/* create a striped file at the temp file path */
mfu_stripe_set(temp_path, stripe_size, stripes);
/* update our status for file create progress */
create_prog_count++;
mfu_progress_update(&create_prog_count, create_prog);
}
/* finalize file create progress messages */
mfu_progress_complete(&create_prog_count, &create_prog);
MPI_Barrier(MPI_COMM_WORLD);
/* initialize progress messages while copying data */
stripe_prog_bytes = 0;
stripe_prog = mfu_progress_start(mfu_progress_timeout, 1, MPI_COMM_WORLD, stripe_progress_fn);
/* found a suffix, now we need to break our files into chunks based on stripe size */
mfu_file_chunk* file_chunks = mfu_file_chunk_list_alloc(filtered, stripe_size);
mfu_file_chunk* p = file_chunks;
while (p != NULL) {
/* build path to temp file */
char temp_path[PATH_MAX];
strcpy(temp_path, p->name);
strcat(temp_path, suffix);
/* write each chunk in our list */
write_file_chunk(p, temp_path);
/* move on to next file chunk */
p = p->next;
}
mfu_file_chunk_list_free(&file_chunks);
/* finalize progress messages */
mfu_progress_complete(&stripe_prog_bytes, &stripe_prog);
MPI_Barrier(MPI_COMM_WORLD);
/* remove input file and rename temp file */
for (idx = 0; idx < size; idx++) {
/* build path to temp file */
const char *in_path = mfu_flist_file_get_name(filtered, idx);
char out_path[PATH_MAX];
strcpy(out_path, in_path);
strcat(out_path, suffix);
/* change the mode of the newly restriped file to be the same as the old one */
mode_t mode = (mode_t) mfu_flist_file_get_mode(filtered, idx);
if (mfu_chmod(out_path, mode) != 0) {
MFU_LOG(MFU_LOG_ERR, "Failed to chmod file %s (%s)", out_path, strerror(errno));
MPI_Abort(MPI_COMM_WORLD, 1);
}
/* rename the new, restriped file to the old name */
if (rename(out_path, in_path) != 0) {
MFU_LOG(MFU_LOG_ERR, "Failed to rename file %s to %s", out_path, in_path);
MPI_Abort(MPI_COMM_WORLD, 1);
}
}
/* wait for everyone to finish */
MPI_Barrier(MPI_COMM_WORLD);
/* free the walk options */
mfu_walk_opts_delete(&walk_opts);
/* free filtered list, path parameters */
mfu_flist_free(&filtered);
mfu_param_path_free_all(numpaths, paths);
mfu_free(&paths);
mfu_finalize();
MPI_Finalize();
return 0;
}
