static void
S_zap_dead_merge(FilePurger *self, Hash *candidates)
{
IndexManager *manager = self->manager;
Lock *merge_lock = IxManager_Make_Merge_Lock(manager);
Lock_Clear_Stale(merge_lock);
if (!Lock_Is_Locked(merge_lock)) {
Hash *merge_data = IxManager_Read_Merge_Data(manager);
Obj *cutoff = merge_data
? Hash_Fetch_Str(merge_data, "cutoff", 6)
: NULL;
if (cutoff) {
CharBuf *cutoff_seg = Seg_num_to_name(Obj_To_I64(cutoff));
if (Folder_Exists(self->folder, cutoff_seg)) {
ZombieCharBuf *merge_json = ZCB_WRAP_STR("merge.json", 10);
DirHandle *dh = Folder_Open_Dir(self->folder, cutoff_seg);
CharBuf *entry = dh ? DH_Get_Entry(dh) : NULL;
CharBuf *filepath = CB_new(32);
if (!dh) {
THROW(ERR, "Can't open segment dir '%o'", filepath);
}
Hash_Store(candidates, (Obj*)cutoff_seg, INCREF(&EMPTY));
Hash_Store(candidates, (Obj*)merge_json, INCREF(&EMPTY));
while (DH_Next(dh)) {
// TODO: recursively delete subdirs within seg dir.
CB_setf(filepath, "%o/%o", cutoff_seg, entry);
Hash_Store(candidates, (Obj*)filepath, INCREF(&EMPTY));
}
DECREF(filepath);
DECREF(dh);
}
DECREF(cutoff_seg);
}
DECREF(merge_data);
}
DECREF(merge_lock);
return;
}
static void
S_zap_dead_merge(FilePurger *self, Hash *candidates) {
FilePurgerIVARS *const ivars = FilePurger_IVARS(self);
IndexManager *manager = ivars->manager;
Lock *merge_lock = IxManager_Make_Merge_Lock(manager);
Lock_Clear_Stale(merge_lock);
if (!Lock_Is_Locked(merge_lock)) {
Hash *merge_data = IxManager_Read_Merge_Data(manager);
Obj *cutoff = merge_data
? Hash_Fetch_Utf8(merge_data, "cutoff", 6)
: NULL;
if (cutoff) {
String *cutoff_seg = Seg_num_to_name(Json_obj_to_i64(cutoff));
if (Folder_Exists(ivars->folder, cutoff_seg)) {
String *merge_json = SSTR_WRAP_UTF8("merge.json", 10);
DirHandle *dh = Folder_Open_Dir(ivars->folder, cutoff_seg);
if (!dh) {
THROW(ERR, "Can't open segment dir '%o'", cutoff_seg);
}
Hash_Store(candidates, cutoff_seg, (Obj*)CFISH_TRUE);
Hash_Store(candidates, merge_json, (Obj*)CFISH_TRUE);
while (DH_Next(dh)) {
// TODO: recursively delete subdirs within seg dir.
String *entry = DH_Get_Entry(dh);
String *filepath = Str_newf("%o/%o", cutoff_seg, entry);
Hash_Store(candidates, filepath, (Obj*)CFISH_TRUE);
DECREF(filepath);
DECREF(entry);
}
DECREF(dh);
}
DECREF(cutoff_seg);
}
DECREF(merge_data);
}
DECREF(merge_lock);
return;
}
Indexer*
Indexer_init(Indexer *self, Schema *schema, Obj *index,
IndexManager *manager, int32_t flags)
{
bool_t create = (flags & Indexer_CREATE) ? true : false;
bool_t truncate = (flags & Indexer_TRUNCATE) ? true : false;
Folder *folder = S_init_folder(index, create);
Lock *write_lock;
CharBuf *latest_snapfile;
Snapshot *latest_snapshot = Snapshot_new();
// Init.
self->stock_doc = Doc_new(NULL, 0);
self->truncate = false;
self->optimize = false;
self->prepared = false;
self->needs_commit = false;
self->snapfile = NULL;
self->merge_lock = NULL;
// Assign.
self->folder = folder;
self->manager = manager
? (IndexManager*)INCREF(manager)
: IxManager_new(NULL, NULL);
IxManager_Set_Folder(self->manager, folder);
// Get a write lock for this folder.
write_lock = IxManager_Make_Write_Lock(self->manager);
Lock_Clear_Stale(write_lock);
if (Lock_Obtain(write_lock)) {
// Only assign if successful, otherwise DESTROY unlocks -- bad!
self->write_lock = write_lock;
}
else {
DECREF(write_lock);
DECREF(self);
RETHROW(INCREF(Err_get_error()));
}
// Find the latest snapshot or create a new one.
latest_snapfile = IxFileNames_latest_snapshot(folder);
if (latest_snapfile) {
Snapshot_Read_File(latest_snapshot, folder, latest_snapfile);
}
// Look for an existing Schema if one wasn't supplied.
if (schema) {
self->schema = (Schema*)INCREF(schema);
}
else {
if (!latest_snapfile) {
THROW(ERR, "No Schema supplied, and can't find one in the index");
}
else {
CharBuf *schema_file = S_find_schema_file(latest_snapshot);
Hash *dump = (Hash*)Json_slurp_json(folder, schema_file);
if (dump) { // read file successfully
self->schema = (Schema*)CERTIFY(
VTable_Load_Obj(SCHEMA, (Obj*)dump), SCHEMA);
schema = self->schema;
DECREF(dump);
schema_file = NULL;
}
else {
THROW(ERR, "Failed to parse %o", schema_file);
}
}
}
// If we're clobbering, start with an empty Snapshot and an empty
// PolyReader. Otherwise, start with the most recent Snapshot and an
// up-to-date PolyReader.
if (truncate) {
self->snapshot = Snapshot_new();
self->polyreader = PolyReader_new(schema, folder, NULL, NULL, NULL);
self->truncate = true;
}
else {
// TODO: clone most recent snapshot rather than read it twice.
self->snapshot = (Snapshot*)INCREF(latest_snapshot);
self->polyreader = latest_snapfile
? PolyReader_open((Obj*)folder, NULL, NULL)
: PolyReader_new(schema, folder, NULL, NULL, NULL);
if (latest_snapfile) {
// Make sure than any existing fields which may have been
// dynamically added during past indexing sessions get added.
Schema *old_schema = PolyReader_Get_Schema(self->polyreader);
Schema_Eat(schema, old_schema);
}
}
// Zap detritus from previous sessions.
{
// Note: we have to feed FilePurger with the most recent snapshot file
// now, but with the Indexer's snapshot later.
FilePurger *file_purger
= FilePurger_new(folder, latest_snapshot, self->manager);
FilePurger_Purge(file_purger);
DECREF(file_purger);
}
// Create a new segment.
{
int64_t new_seg_num
= IxManager_Highest_Seg_Num(self->manager, latest_snapshot) + 1;
Lock *merge_lock = IxManager_Make_Merge_Lock(self->manager);
uint32_t i, max;
if (Lock_Is_Locked(merge_lock)) {
// If there's a background merge process going on, stay out of its
// way.
Hash *merge_data = IxManager_Read_Merge_Data(self->manager);
Obj *cutoff_obj = merge_data
? Hash_Fetch_Str(merge_data, "cutoff", 6)
: NULL;
if (!cutoff_obj) {
DECREF(merge_lock);
DECREF(merge_data);
THROW(ERR, "Background merge detected, but can't read merge data");
}
else {
int64_t cutoff = Obj_To_I64(cutoff_obj);
if (cutoff >= new_seg_num) {
new_seg_num = cutoff + 1;
}
}
DECREF(merge_data);
}
self->segment = Seg_new(new_seg_num);
// Add all known fields to Segment.
{
VArray *fields = Schema_All_Fields(schema);
for (i = 0, max = VA_Get_Size(fields); i < max; i++) {
Seg_Add_Field(self->segment, (CharBuf*)VA_Fetch(fields, i));
}
DECREF(fields);
}
DECREF(merge_lock);
}
// Create new SegWriter and FilePurger.
self->file_purger
= FilePurger_new(folder, self->snapshot, self->manager);
self->seg_writer = SegWriter_new(self->schema, self->snapshot,
self->segment, self->polyreader);
SegWriter_Prep_Seg_Dir(self->seg_writer);
// Grab a local ref to the DeletionsWriter.
self->del_writer = (DeletionsWriter*)INCREF(
SegWriter_Get_Del_Writer(self->seg_writer));
DECREF(latest_snapfile);
DECREF(latest_snapshot);
return self;
}
static bool_t
S_maybe_merge(Indexer *self, VArray *seg_readers)
{
bool_t merge_happened = false;
uint32_t num_seg_readers = VA_Get_Size(seg_readers);
Lock *merge_lock = IxManager_Make_Merge_Lock(self->manager);
bool_t got_merge_lock = Lock_Obtain(merge_lock);
int64_t cutoff;
VArray *to_merge;
uint32_t i, max;
if (got_merge_lock) {
self->merge_lock = merge_lock;
cutoff = 0;
}
else {
// If something else holds the merge lock, don't interfere.
Hash *merge_data = IxManager_Read_Merge_Data(self->manager);
if (merge_data) {
Obj *cutoff_obj = Hash_Fetch_Str(merge_data, "cutoff", 6);
if (cutoff_obj) {
cutoff = Obj_To_I64(cutoff_obj);
}
else {
cutoff = I64_MAX;
}
DECREF(merge_data);
}
else {
cutoff = I64_MAX;
}
DECREF(merge_lock);
}
// Get a list of segments to recycle. Validate and confirm that there are
// no dupes in the list.
to_merge = IxManager_Recycle(self->manager, self->polyreader,
self->del_writer, cutoff, self->optimize);
{
Hash *seen = Hash_new(VA_Get_Size(to_merge));
for (i = 0, max = VA_Get_Size(to_merge); i < max; i++) {
SegReader *seg_reader = (SegReader*)CERTIFY(
VA_Fetch(to_merge, i), SEGREADER);
CharBuf *seg_name = SegReader_Get_Seg_Name(seg_reader);
if (Hash_Fetch(seen, (Obj*)seg_name)) {
DECREF(seen);
DECREF(to_merge);
THROW(ERR, "Recycle() tried to merge segment '%o' twice",
seg_name);
}
Hash_Store(seen, (Obj*)seg_name, INCREF(&EMPTY));
}
DECREF(seen);
}
// Consolidate segments if either sparse or optimizing forced.
for (i = 0, max = VA_Get_Size(to_merge); i < max; i++) {
SegReader *seg_reader = (SegReader*)VA_Fetch(to_merge, i);
int64_t seg_num = SegReader_Get_Seg_Num(seg_reader);
Matcher *deletions
= DelWriter_Seg_Deletions(self->del_writer, seg_reader);
I32Array *doc_map = DelWriter_Generate_Doc_Map(self->del_writer,
deletions, SegReader_Doc_Max(seg_reader),
(int32_t)Seg_Get_Count(self->segment)
);
if (seg_num <= cutoff) {
THROW(ERR, "Segment %o violates cutoff (%i64 <= %i64)",
SegReader_Get_Seg_Name(seg_reader), seg_num, cutoff);
}
SegWriter_Merge_Segment(self->seg_writer, seg_reader, doc_map);
merge_happened = true;
DECREF(deletions);
DECREF(doc_map);
}
// Write out new deletions.
if (DelWriter_Updated(self->del_writer)) {
// Only write out if they haven't all been applied.
if (VA_Get_Size(to_merge) != num_seg_readers) {
DelWriter_Finish(self->del_writer);
}
}
DECREF(to_merge);
return merge_happened;
}