BackgroundMerger*
BGMerger_init(BackgroundMerger *self, Obj *index, IndexManager *manager) {
BackgroundMergerIVARS *const ivars = BGMerger_IVARS(self);
Folder *folder = S_init_folder(index);
// Init.
ivars->optimize = false;
ivars->prepared = false;
ivars->needs_commit = false;
ivars->snapfile = NULL;
ivars->doc_maps = Hash_new(0);
// Assign.
ivars->folder = folder;
if (manager) {
ivars->manager = (IndexManager*)INCREF(manager);
}
else {
ivars->manager = IxManager_new(NULL, NULL);
IxManager_Set_Write_Lock_Timeout(ivars->manager, 10000);
}
IxManager_Set_Folder(ivars->manager, folder);
// Obtain write lock (which we'll only hold briefly), then merge lock.
S_obtain_write_lock(self);
if (!ivars->write_lock) {
DECREF(self);
RETHROW(INCREF(Err_get_error()));
}
S_obtain_merge_lock(self);
if (!ivars->merge_lock) {
DECREF(self);
RETHROW(INCREF(Err_get_error()));
}
// Find the latest snapshot. If there's no index content, bail early.
ivars->snapshot = Snapshot_Read_File(Snapshot_new(), folder, NULL);
if (!Snapshot_Get_Path(ivars->snapshot)) {
S_release_write_lock(self);
S_release_merge_lock(self);
return self;
}
// Create FilePurger. Zap detritus from previous sessions.
ivars->file_purger = FilePurger_new(folder, ivars->snapshot, ivars->manager);
FilePurger_Purge(ivars->file_purger);
// Open a PolyReader, passing in the IndexManager so we get a read lock on
// the Snapshot's files -- so that Indexers don't zap our files while
// we're operating in the background.
ivars->polyreader = PolyReader_open((Obj*)folder, NULL, ivars->manager);
// Clone the PolyReader's schema.
Obj *dump = (Obj*)Schema_Dump(PolyReader_Get_Schema(ivars->polyreader));
ivars->schema = (Schema*)CERTIFY(Freezer_load(dump), SCHEMA);
DECREF(dump);
// Create new Segment.
int64_t new_seg_num
= IxManager_Highest_Seg_Num(ivars->manager, ivars->snapshot) + 1;
Vector *fields = Schema_All_Fields(ivars->schema);
ivars->segment = Seg_new(new_seg_num);
for (uint32_t i = 0, max = Vec_Get_Size(fields); i < max; i++) {
Seg_Add_Field(ivars->segment, (String*)Vec_Fetch(fields, i));
}
DECREF(fields);
// Our "cutoff" is the segment this BackgroundMerger will write. Now that
// we've determined the cutoff, write the merge data file.
ivars->cutoff = Seg_Get_Number(ivars->segment);
IxManager_Write_Merge_Data(ivars->manager, ivars->cutoff);
/* Create the SegWriter but hold off on preparing the new segment
* directory -- because if we don't need to merge any segments we don't
* need it. (We've reserved the dir by plopping down the merge.json
* file.) */
ivars->seg_writer = SegWriter_new(ivars->schema, ivars->snapshot,
ivars->segment, ivars->polyreader);
// Grab a local ref to the DeletionsWriter.
ivars->del_writer
= (DeletionsWriter*)INCREF(SegWriter_Get_Del_Writer(ivars->seg_writer));
// Release the write lock. Now new Indexers can start while we work in
// the background.
S_release_write_lock(self);
return self;
}
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
S_merge_updated_deletions(BackgroundMerger *self) {
BackgroundMergerIVARS *const ivars = BGMerger_IVARS(self);
Hash *updated_deletions = NULL;
PolyReader *new_polyreader
= PolyReader_open((Obj*)ivars->folder, NULL, NULL);
Vector *new_seg_readers
= PolyReader_Get_Seg_Readers(new_polyreader);
Vector *old_seg_readers
= PolyReader_Get_Seg_Readers(ivars->polyreader);
Hash *new_segs = Hash_new(Vec_Get_Size(new_seg_readers));
for (uint32_t i = 0, max = Vec_Get_Size(new_seg_readers); i < max; i++) {
SegReader *seg_reader = (SegReader*)Vec_Fetch(new_seg_readers, i);
String *seg_name = SegReader_Get_Seg_Name(seg_reader);
Hash_Store(new_segs, seg_name, INCREF(seg_reader));
}
for (uint32_t i = 0, max = Vec_Get_Size(old_seg_readers); i < max; i++) {
SegReader *seg_reader = (SegReader*)Vec_Fetch(old_seg_readers, i);
String *seg_name = SegReader_Get_Seg_Name(seg_reader);
// If this segment was merged away...
if (Hash_Fetch(ivars->doc_maps, seg_name)) {
SegReader *new_seg_reader
= (SegReader*)CERTIFY(
Hash_Fetch(new_segs, seg_name),
SEGREADER);
int32_t old_del_count = SegReader_Del_Count(seg_reader);
int32_t new_del_count = SegReader_Del_Count(new_seg_reader);
// ... were any new deletions applied against it?
if (old_del_count != new_del_count) {
DeletionsReader *del_reader
= (DeletionsReader*)SegReader_Obtain(
new_seg_reader,
Class_Get_Name(DELETIONSREADER));
if (!updated_deletions) {
updated_deletions = Hash_new(max);
}
Hash_Store(updated_deletions, seg_name,
(Obj*)DelReader_Iterator(del_reader));
}
}
}
DECREF(new_polyreader);
DECREF(new_segs);
if (!updated_deletions) {
return false;
}
else {
PolyReader *merge_polyreader
= PolyReader_open((Obj*)ivars->folder, ivars->snapshot, NULL);
Vector *merge_seg_readers
= PolyReader_Get_Seg_Readers(merge_polyreader);
Snapshot *latest_snapshot
= Snapshot_Read_File(Snapshot_new(), ivars->folder, NULL);
int64_t new_seg_num
= IxManager_Highest_Seg_Num(ivars->manager, latest_snapshot) + 1;
Segment *new_segment = Seg_new(new_seg_num);
SegWriter *seg_writer = SegWriter_new(ivars->schema, ivars->snapshot,
new_segment, merge_polyreader);
DeletionsWriter *del_writer = SegWriter_Get_Del_Writer(seg_writer);
int64_t merge_seg_num = Seg_Get_Number(ivars->segment);
uint32_t seg_tick = INT32_MAX;
int32_t offset = INT32_MAX;
SegWriter_Prep_Seg_Dir(seg_writer);
for (uint32_t i = 0, max = Vec_Get_Size(merge_seg_readers); i < max; i++) {
SegReader *seg_reader
= (SegReader*)Vec_Fetch(merge_seg_readers, i);
if (SegReader_Get_Seg_Num(seg_reader) == merge_seg_num) {
I32Array *offsets = PolyReader_Offsets(merge_polyreader);
seg_tick = i;
offset = I32Arr_Get(offsets, seg_tick);
DECREF(offsets);
}
}
if (offset == INT32_MAX) { THROW(ERR, "Failed sanity check"); }
HashIterator *iter = HashIter_new(updated_deletions);
while (HashIter_Next(iter)) {
String *seg_name = HashIter_Get_Key(iter);
Matcher *deletions = (Matcher*)HashIter_Get_Value(iter);
I32Array *doc_map
= (I32Array*)CERTIFY(
Hash_Fetch(ivars->doc_maps, seg_name),
I32ARRAY);
int32_t del;
while (0 != (del = Matcher_Next(deletions))) {
// Find the slot where the deleted doc resides in the
// rewritten segment. If the doc was already deleted when we
// were merging, do nothing.
int32_t remapped = I32Arr_Get(doc_map, del);
if (remapped) {
// It's a new deletion, so carry it forward and zap it in
// the rewritten segment.
DelWriter_Delete_By_Doc_ID(del_writer, remapped + offset);
}
}
}
DECREF(iter);
// Finish the segment and clean up.
DelWriter_Finish(del_writer);
SegWriter_Finish(seg_writer);
DECREF(seg_writer);
DECREF(new_segment);
DECREF(latest_snapshot);
DECREF(merge_polyreader);
DECREF(updated_deletions);
}
return true;
}