/*
* appTeardown()
*
* Teardown all modules, close all files, and tidy up all
* application state.
*
* This function is idempotent.
*/
static void
appTeardown(
void)
{
static int teardown_flag = 0;
if (teardown_flag) {
return;
}
teardown_flag = 1;
skAggBagDestroy(&ab);
/* close output */
skStreamClose(output);
skStreamDestroy(&output);
/* destroy string maps for keys and counters */
skStringMapDestroy(key_name_map);
key_name_map = NULL;
skStringMapDestroy(counter_name_map);
counter_name_map = NULL;
skAggBagOptionsTeardown();
skOptionsCtxDestroy(&optctx);
skAppUnregister();
}
int main(int argc, char **argv)
{
int rv;
appSetup(argc, argv); /* never returns on error */
sortRandom();
/* close the file */
if ((rv = skStreamClose(out_rwios))
|| (rv = skStreamDestroy(&out_rwios)))
{
skStreamPrintLastErr(out_rwios, rv, &skAppPrintErr);
appExit(EXIT_FAILURE);
}
out_rwios = NULL;
appExit(EXIT_SUCCESS);
return 0; /* NOTREACHED */
}
/*
* status = readerGetRecord(&out_rwrec, &out_probe, flow_processor);
*
* Invoked by input_mode_type->get_record_fn();
*/
static fp_get_record_result_t
readerGetRecord(
rwRec *out_rwrec,
const skpc_probe_t **out_probe,
flow_proc_t *fproc)
{
static pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
skstream_t *fcfile;
const char *filename;
fp_get_record_result_t retVal = FP_GET_ERROR;
int rv;
pthread_mutex_lock(&mutex);
/* If we don't have a source, get a file from the directory poller
* and start processing it. */
if (fproc->flow_src == NULL) {
switch (readerGetNextValidFile(fproc)) {
case 0:
/* Success */
break;
case -1:
/* Error getting file name (maybe in shutdown?) */
goto END;
case -2: /* Error opening file */
default: /* Unexpected value */
retVal = FP_FATAL_ERROR;
goto END;
}
}
fcfile = (skstream_t*)fproc->flow_src;
/* Assume we can get a record from the probe. */
retVal = FP_RECORD;
*out_probe = fproc->probe;
/* Try to get a record */
rv = skStreamReadRecord(fcfile, out_rwrec);
if (rv) {
/* get failed: either at EOF or got an error. */
if (rv != SKSTREAM_ERR_EOF) {
skStreamPrintLastErr(fcfile, rv, &WARNINGMSG);
}
retVal = FP_FILE_BREAK;
*out_probe = NULL;
/* Print results for the file we just processed. */
filename = skStreamGetPathname(fcfile);
INFOMSG("Processed file %s, %" PRIu64 " records.",
filename, skStreamGetRecordCount(fcfile));
skStreamClose(fcfile);
/* Either archive the file or remove it */
archiveDirectoryInsertOrRemove(filename, NULL);
/* All done with the flow source */
skStreamDestroy(&fcfile);
fproc->flow_src = NULL;
fproc->probe = NULL;
}
END:
pthread_mutex_unlock(&mutex);
return retVal;
}
/*
* mergeFiles(temp_file_idx)
*
* Merge the temporary files numbered from 0 to 'temp_file_idx'
* inclusive into the output file 'out_ios', maintaining sorted
* order. Exits the application if an error occurs.
*/
static void
mergeFiles(
int temp_file_idx)
{
char errbuf[2 * PATH_MAX];
skstream_t *fps[MAX_MERGE_FILES];
uint8_t recs[MAX_MERGE_FILES][NODE_SIZE];
uint8_t lowest_rec[NODE_SIZE];
int j;
uint16_t open_count;
uint16_t i;
uint16_t lowest;
uint16_t *top_heap;
int tmp_idx_a;
int tmp_idx_b;
skstream_t *fp_intermediate = NULL;
int tmp_idx_intermediate;
skheap_t *heap;
uint32_t heap_count;
int opened_all_temps = 0;
ssize_t rv;
/* the index of the first temp file to the merge */
tmp_idx_a = 0;
TRACEMSG(("Merging #%d through #%d to '%s'",
tmp_idx_a, temp_file_idx, skStreamGetPathname(out_rwios)));
heap = skHeapCreate2(compHeapNodes, MAX_MERGE_FILES, sizeof(uint16_t),
NULL, recs);
if (NULL == heap) {
skAppPrintOutOfMemory("heap");
appExit(EXIT_FAILURE);
}
/* This loop repeats as long as we haven't read all of the temp
* files generated in the qsort stage. */
do {
assert(SKHEAP_ERR_EMPTY==skHeapPeekTop(heap,(skheapnode_t*)&top_heap));
/* the index of the list temp file to merge */
tmp_idx_b = temp_file_idx;
/* open an intermediate temp file. The merge-sort will have
* to write records here if there are not enough file handles
* available to open all the existing tempoary files. */
fp_intermediate = skTempFileCreateStream(tmpctx, &tmp_idx_intermediate);
if (fp_intermediate == NULL) {
skAppPrintSyserror("Error creating new temporary file");
appExit(EXIT_FAILURE);
}
/* count number of files we open */
open_count = 0;
/* Attempt to open up to MAX_MERGE_FILES, though we an open
* may fail due to lack of resources (EMFILE or ENOMEM) */
for (j = tmp_idx_a; j <= tmp_idx_b; ++j) {
fps[open_count] = skTempFileOpenStream(tmpctx, j);
if (fps[open_count] == NULL) {
if ((open_count > 0)
&& ((errno == EMFILE) || (errno == ENOMEM)))
{
/* We cannot open any more files. Rewind counter
* by one to catch this file on the next merge */
tmp_idx_b = j - 1;
TRACEMSG((("FILE limit hit--"
"merging #%d through #%d into #%d: %s"),
tmp_idx_a, tmp_idx_b, tmp_idx_intermediate,
strerror(errno)));
break;
} else {
skAppPrintSyserror(("Error opening existing"
" temporary file '%s'"),
skTempFileGetName(tmpctx, j));
appExit(EXIT_FAILURE);
}
}
/* read the first record */
rv = skStreamRead(fps[open_count], recs[open_count], NODE_SIZE);
if (NODE_SIZE == rv) {
/* insert the file index into the heap */
skHeapInsert(heap, &open_count);
++open_count;
if (open_count == MAX_MERGE_FILES) {
/* We've reached the limit for this pass. Set
* tmp_idx_b to the file we just opened. */
tmp_idx_b = j;
TRACEMSG((("MAX_MERGE_FILES limit hit--"
"merging #%d through #%d to #%d"),
tmp_idx_a, tmp_idx_b, tmp_idx_intermediate));
break;
}
} else if (0 == rv) {
TRACEMSG(("Ignoring empty temporary file '%s'",
skTempFileGetName(tmpctx, j)));
skStreamDestroy(&fps[open_count]);
} else {
if (rv > 0) {
snprintf(errbuf, sizeof(errbuf),
"Short read %" SK_PRIdZ "/%" PRIu32 " from '%s'",
rv, NODE_SIZE,
skStreamGetPathname(fps[open_count]));
} else {
skStreamLastErrMessage(
fps[open_count], rv, errbuf, sizeof(errbuf));
}
skAppPrintErr(
"Error reading first record from temporary file: %s",
errbuf);
appExit(EXIT_FAILURE);
}
}
/* Here, we check to see if we've opened all temp files. If
* so, set a flag so we write data to final destination and
* break out of the loop after we're done. */
if (tmp_idx_b == temp_file_idx) {
opened_all_temps = 1;
/* no longer need the intermediate temp file */
skStreamDestroy(&fp_intermediate);
} else {
/* we could not open all temp files, so merge all opened
* temp files into the intermediate file. Add the
* intermediate file to the list of files to merge */
temp_file_idx = tmp_idx_intermediate;
}
TRACEMSG((("Merging %" PRIu16 " temporary files"), open_count));
heap_count = skHeapGetNumberEntries(heap);
assert(heap_count == open_count);
/* get the index of the file with the lowest record; which is
* at the top of the heap */
if (skHeapPeekTop(heap, (skheapnode_t*)&top_heap) != SKHEAP_OK) {
skAppPrintErr("Unable to open and read any temporary files.");
appExit(EXIT_FAILURE);
}
lowest = *top_heap;
/* exit this do...while() once all records for all opened
* files have been read */
do {
/* lowest_rec is the record pointed to by the index at the
* top of the heap */
memcpy(lowest_rec, recs[lowest], NODE_SIZE);
/* write the record */
if (fp_intermediate) {
/* write the record to intermediate tmp file */
rv = skStreamWrite(fp_intermediate, lowest_rec, NODE_SIZE);
if (NODE_SIZE != rv) {
skAppPrintSyserror(
"Error writing record to temporary file '%s'",
skTempFileGetName(tmpctx, tmp_idx_intermediate));
appExit(EXIT_FAILURE);
}
} else {
/* we successfully opened all (remaining) temp files,
* write to record to the final destination */
rv = skStreamWriteRecord(out_rwios, (rwRec*)lowest_rec);
if (0 != rv) {
skStreamPrintLastErr(out_rwios, rv, &skAppPrintErr);
if (SKSTREAM_ERROR_IS_FATAL(rv)) {
appExit(EXIT_FAILURE);
}
}
}
/* replace the record we just processed and loop over all
* files until we get a record that is not a duplicate */
do {
if ((rv = skStreamRead(fps[lowest], recs[lowest], NODE_SIZE))
!= NODE_SIZE)
{
/* read failed. there is no more data for this
* file; remove it from the heap; if the heap is
* empty, exit the loop */
skHeapExtractTop(heap, NULL);
--heap_count;
#if TRACEMSG_LEVEL > 0
if (rv == 0) {
TRACEMSG(
("Finished reading file #%u: EOF; %u files remain",
(tmp_idx_a + lowest), heap_count));
} else if (rv > 0) {
TRACEMSG(
("Finished reading file #%u: Short read "
"%" SK_PRIdZ "/%" PRIu32 "; %u files remain",
tmp_idx_a + lowest, rv, NODE_SIZE, heap_count));
} else {
skStreamLastErrMessage(
fps[open_count], rv, errbuf, sizeof(errbuf));
TRACEMSG(
("Finished reading file #%u: %s; %u files remain",
(tmp_idx_a + lowest), errbuf, heap_count));
}
#endif /* TRACEMSG_LEVEL */
if (0 == heap_count) {
break;
}
} else if (rwrecCompare(lowest_rec, recs[lowest])) {
/* read succeeded. new record is not a
* duplicate and we insert it into the heap */
/* FIXME: This comparison reduces work when the
* keys are the same, but it adds another
* comparison when the keys are different; is this
* an overall win or lose? */
skHeapReplaceTop(heap, &lowest, NULL);
} else {
/* read succeeded. record is a duplicate; ignore
* the record and leave the heap unchanged */
continue;
}
/* get the record at the top of the heap and see if it
* is a duplicate; if it is, ignore it. */
skHeapPeekTop(heap, (skheapnode_t*)&top_heap);
lowest = *top_heap;
} while (0 == rwrecCompare(lowest_rec, recs[lowest]));
} while (heap_count > 0);
TRACEMSG((("Finished processing #%d through #%d"),
tmp_idx_a, tmp_idx_b));
/* Close all open temp files */
for (i = 0; i < open_count; ++i) {
skStreamDestroy(&fps[i]);
}
/* Delete all temp files we opened (or attempted to open) this
* time */
for (j = tmp_idx_a; j <= tmp_idx_b; ++j) {
skTempFileRemove(tmpctx, j);
}
/* Close the intermediate temp file. */
if (fp_intermediate) {
rv = skStreamClose(fp_intermediate);
if (rv) {
skStreamLastErrMessage(
fp_intermediate, rv, errbuf, sizeof(errbuf));
skAppPrintErr("Error closing temporary file: %s", errbuf);
skStreamDestroy(&fp_intermediate);
appExit(EXIT_FAILURE);
}
skStreamDestroy(&fp_intermediate);
}
/* Start the next merge with the next input temp file */
tmp_idx_a = tmp_idx_b + 1;
} while (!opened_all_temps);
skHeapFree(heap);
}