/*
* InitCatalogCache - initialize the caches
*
* Note that no database access is done here; we only allocate memory
* and initialize the cache structure. Interrogation of the database
* to complete initialization of a cache happens upon first use
* of that cache.
*/
void
InitCatalogCache(void)
{
int cacheId;
int i,
j;
StaticAssertStmt(SysCacheSize == (int) lengthof(cacheinfo),
"SysCacheSize does not match syscache.c's array");
Assert(!CacheInitialized);
SysCacheRelationOidSize = SysCacheSupportingRelOidSize = 0;
for (cacheId = 0; cacheId < SysCacheSize; cacheId++)
{
SysCache[cacheId] = InitCatCache(cacheId,
cacheinfo[cacheId].reloid,
cacheinfo[cacheId].indoid,
cacheinfo[cacheId].nkeys,
cacheinfo[cacheId].key,
cacheinfo[cacheId].nbuckets);
if (!PointerIsValid(SysCache[cacheId]))
elog(ERROR, "could not initialize cache %u (%d)",
cacheinfo[cacheId].reloid, cacheId);
/* Accumulate data for OID lists, too */
SysCacheRelationOid[SysCacheRelationOidSize++] =
cacheinfo[cacheId].reloid;
SysCacheSupportingRelOid[SysCacheSupportingRelOidSize++] =
cacheinfo[cacheId].reloid;
SysCacheSupportingRelOid[SysCacheSupportingRelOidSize++] =
cacheinfo[cacheId].indoid;
/* see comments for RelationInvalidatesSnapshotsOnly */
Assert(!RelationInvalidatesSnapshotsOnly(cacheinfo[cacheId].reloid));
}
Assert(SysCacheRelationOidSize <= lengthof(SysCacheRelationOid));
Assert(SysCacheSupportingRelOidSize <= lengthof(SysCacheSupportingRelOid));
/* Sort and de-dup OID arrays, so we can use binary search. */
pg_qsort(SysCacheRelationOid, SysCacheRelationOidSize,
sizeof(Oid), oid_compare);
for (i = 1, j = 0; i < SysCacheRelationOidSize; i++)
{
if (SysCacheRelationOid[i] != SysCacheRelationOid[j])
SysCacheRelationOid[++j] = SysCacheRelationOid[i];
}
SysCacheRelationOidSize = j + 1;
pg_qsort(SysCacheSupportingRelOid, SysCacheSupportingRelOidSize,
sizeof(Oid), oid_compare);
for (i = 1, j = 0; i < SysCacheSupportingRelOidSize; i++)
{
if (SysCacheSupportingRelOid[i] != SysCacheSupportingRelOid[j])
SysCacheSupportingRelOid[++j] = SysCacheSupportingRelOid[i];
}
SysCacheSupportingRelOidSize = j + 1;
CacheInitialized = true;
}
/*
* InitCatalogCache - initialize the caches
*
* Note that no database access is done here; we only allocate memory
* and initialize the cache structure. Interrogation of the database
* to complete initialization of a cache happens upon first use
* of that cache.
*/
void
InitCatalogCache(void)
{
int cacheId;
int i,
j = 0;
Assert(!CacheInitialized);
MemSet(SysCache, 0, sizeof(SysCache));
for (cacheId = 0; cacheId < SysCacheSize; cacheId++)
{
SysCache[cacheId] = InitCatCache(cacheId,
cacheinfo[cacheId].reloid,
cacheinfo[cacheId].indoid,
cacheinfo[cacheId].nkeys,
cacheinfo[cacheId].key,
cacheinfo[cacheId].nbuckets);
if (!PointerIsValid(SysCache[cacheId]))
elog(ERROR, "could not initialize cache %u (%d)",
cacheinfo[cacheId].reloid, cacheId);
SysCacheRelationOid[SysCacheRelationOidSize++] =
cacheinfo[cacheId].reloid;
/* see comments for RelationInvalidatesSnapshotsOnly */
Assert(!RelationInvalidatesSnapshotsOnly(cacheinfo[cacheId].reloid));
}
/* Sort and dedup OIDs. */
pg_qsort(SysCacheRelationOid, SysCacheRelationOidSize,
sizeof(Oid), oid_compare);
for (i = 1; i < SysCacheRelationOidSize; ++i)
if (SysCacheRelationOid[i] != SysCacheRelationOid[j])
SysCacheRelationOid[++j] = SysCacheRelationOid[i];
SysCacheRelationOidSize = j + 1;
CacheInitialized = true;
}
static void
SaveBuffers(void)
{
int i;
int num_buffers;
int log_level = DEBUG3;
SavedBuffer *saved_buffers;
volatile BufferDesc *bufHdr; // XXX: Do we really need volatile here?
FILE *file = NULL;
int database_counter= 0;
Oid prev_database = InvalidOid;
Oid prev_filenode = InvalidOid;
ForkNumber prev_forknum = InvalidForkNumber;
BlockNumber prev_blocknum = InvalidBlockNumber;
BlockNumber range_counter = 0;
const char *savefile_path;
/*
* XXX: If the memory request fails, ask for a smaller memory chunk, and use
* it to create chunks of save-files, and make the workers read those chunks.
*
* This is not a concern as of now, so deferred; there's at least one other
* place that allocates (NBuffers * (much_bigger_struct)), so this seems to
* be an acceptable practice.
*/
saved_buffers = (SavedBuffer *) palloc(sizeof(SavedBuffer) * NBuffers);
/* Lock the buffer partitions for reading. */
for (i = 0; i < NUM_BUFFER_PARTITIONS; ++i)
LWLockAcquire(FirstBufMappingLock + i, LW_SHARED);
/* Scan and save a list of valid buffers. */
for (num_buffers = 0, i = 0, bufHdr = BufferDescriptors; i < NBuffers; ++i, ++bufHdr)
{
/* Lock each buffer header before inspecting. */
LockBufHdr(bufHdr);
/* Skip invalid buffers */
if ((bufHdr->flags & BM_VALID) && (bufHdr->flags & BM_TAG_VALID))
{
saved_buffers[num_buffers].database = bufHdr->tag.rnode.dbNode;
saved_buffers[num_buffers].filenode = bufHdr->tag.rnode.relNode;
saved_buffers[num_buffers].forknum = bufHdr->tag.forkNum;
saved_buffers[num_buffers].blocknum = bufHdr->tag.blockNum;
++num_buffers;
}
UnlockBufHdr(bufHdr);
}
/* Unlock the buffer partitions in reverse order, to avoid a deadlock. */
for (i = NUM_BUFFER_PARTITIONS - 1; i >= 0; --i)
LWLockRelease(FirstBufMappingLock + i);
/*
* Sort the list, so that we can optimize the storage of these buffers.
*
* The side-effect of this storage optimization is that when reading the
* blocks back from relation forks, it leads to sequential reads, which
* improve the restore speeds quite considerably as compared to random reads
* from different blocks all over the data directory.
*/
pg_qsort(saved_buffers, num_buffers, sizeof(SavedBuffer), SavedBufferCmp);
/* Connect to the database and start a transaction for database name lookups. */
BackgroundWorkerInitializeConnection(guc_default_database, NULL);
SetCurrentStatementStartTimestamp();
StartTransactionCommand();
PushActiveSnapshot(GetTransactionSnapshot());
pgstat_report_activity(STATE_RUNNING, "saving buffers");
for (i = 0; i < num_buffers; ++i)
{
int j;
SavedBuffer *buf = &saved_buffers[i];
if (i == 0)
{
/*
* Special case for global objects. The sort brings them to the
* front of the list.
*/
/* Make sure the first buffer we save belongs to global object. */
Assert(buf->database == InvalidOid);
/*
* Database number (and save-file name) 1 is reserverd for storing
* list of buffers of global objects.
*/
database_counter = 1;
savefile_path = getSavefileName(database_counter);
file = fileOpen(savefile_path, PG_BINARY_W);
writeDBName("", file, savefile_path);
prev_database = buf->database;
}
if (buf->database != prev_database)
{
char *dbname;
/*
* We are beginning to process a different database than the
* previous one; close the save-file of previous database, and open
* a new one.
*/
++database_counter;
dbname = get_database_name(buf->database);
Assert(dbname != NULL);
if (file != NULL)
fileClose(file, savefile_path);
savefile_path = getSavefileName(database_counter);
file = fileOpen(savefile_path, PG_BINARY_W);
writeDBName(dbname, file, savefile_path);
pfree(dbname);
/* Reset trackers appropriately */
prev_database = buf->database;
prev_filenode = InvalidOid;
prev_forknum = InvalidForkNumber;
prev_blocknum = InvalidBlockNumber;
range_counter = 0;
}
if (buf->filenode != prev_filenode)
{
/* We're beginning to process a new relation; emit a record for it. */
fileWrite("r", 1, file, savefile_path);
fileWrite(&(buf->filenode), sizeof(Oid), file, savefile_path);
/* Reset trackers appropriately */
prev_filenode = buf->filenode;
prev_forknum = InvalidForkNumber;
prev_blocknum = InvalidBlockNumber;
range_counter = 0;
}
if (buf->forknum != prev_forknum)
{
/*
* We're beginning to process a new fork of this relation; add a
* record for it.
*/
fileWrite("f", 1, file, savefile_path);
fileWrite(&(buf->forknum), sizeof(ForkNumber), file, savefile_path);
/* Reset trackers appropriately */
prev_forknum = buf->forknum;
prev_blocknum = InvalidBlockNumber;
range_counter = 0;
}
ereport(log_level,
(errmsg("writer: writing block db %d filenode %d forknum %d blocknum %d",
database_counter, prev_filenode, prev_forknum, buf->blocknum)));
fileWrite("b", 1, file, savefile_path);
fileWrite(&(buf->blocknum), sizeof(BlockNumber), file, savefile_path);
prev_blocknum = buf->blocknum;
/*
* If a continuous range of blocks follows this block, then emit one
* entry for the range, instead of one for each block.
*/
range_counter = 0;
for ( j = i+1; j < num_buffers; ++j)
{
SavedBuffer *tmp = &saved_buffers[j];
if (tmp->database == prev_database
&& tmp->filenode == prev_filenode
&& tmp->forknum == prev_forknum
&& tmp->blocknum == (prev_blocknum + range_counter + 1))
{
++range_counter;
}
}
if (range_counter != 0)
{
ereport(log_level,
(errmsg("writer: writing range db %d filenode %d forknum %d blocknum %d range %d",
database_counter, prev_filenode, prev_forknum, prev_blocknum, range_counter)));
fileWrite("N", 1, file, savefile_path);
fileWrite(&range_counter, sizeof(range_counter), file, savefile_path);
i += range_counter;
}
}
ereport(LOG,
(errmsg("Buffer Saver: saved metadata of %d blocks", num_buffers)));
Assert(file != NULL);
fileClose(file, savefile_path);
pfree(saved_buffers);
PopActiveSnapshot();
CommitTransactionCommand();
pgstat_report_activity(STATE_IDLE, NULL);
}
