/*
* Open an Append Only relation file segment
*
* The fd module's PathNameOpenFile() is used to open the file, so the
* the File* routines can be used to read, write, close, etc, the file.
*/
bool
OpenAOSegmentFile(
Relation rel,
char *filepathname,
int32 segmentFileNum,
int64 logicalEof,
MirroredAppendOnlyOpen *mirroredOpen)
{
ItemPointerData persistentTid;
int64 persistentSerialNum;
int primaryError;
if (!ReadGpRelationNode(
rel->rd_rel->reltablespace,
rel->rd_rel->relfilenode,
segmentFileNum,
&persistentTid,
&persistentSerialNum))
{
if (logicalEof == 0)
return false;
elog(ERROR, "Did not find gp_relation_node entry for relation name %s, relation id %u, relfilenode %u, segment file #%d, logical eof " INT64_FORMAT,
rel->rd_rel->relname.data,
rel->rd_id,
rel->rd_node.relNode,
segmentFileNum,
logicalEof);
}
MirroredAppendOnly_OpenReadWrite(
mirroredOpen,
&rel->rd_node,
segmentFileNum,
/* relationName */ NULL, // Ok to be NULL -- we don't know the name here.
logicalEof,
/* traceOpenFlags */ false,
&persistentTid,
persistentSerialNum,
&primaryError);
if (primaryError != 0)
ereport(ERROR,
(errcode_for_file_access(),
errmsg("Could not open Append-Only segment file '%s': %s",
filepathname,
strerror(primaryError))));
return true;
}
/*
* Opens the next segment file to write. The file must already exist.
* This routine is responsible for seeking to the proper write location
* given the logical EOF.
*
* @filePathName: The name of the segment file to open.
* @logicalEof: The last committed write transaction's EOF
* value to use as the end of the segment file.
* @parquet_file The file handler of segment file
*/
static void OpenSegmentFile(
MirroredAppendOnlyOpen *mirroredOpen,
char *filePathName,
int64 logicalEof,
RelFileNode *relFileNode,
int32 segmentFileNum,
char *relname,
File *parquet_file,
File *parquet_file_previous,
CompactProtocol **protocol_read,
TupleDesc tableAttrs,
ParquetMetadata *parquetMetadata,
int64 *fileLen,
int64 *fileLen_uncompressed,
int *previous_rowgroupcnt) {
int primaryError;
File file;
int64 seekResult;
Assert(filePathName != NULL);
bool metadataExist = false;
/*
* Open the file for metadata reading.
*/
MirroredAppendOnly_OpenReadWrite(mirroredOpen, relFileNode, segmentFileNum,
relname, logicalEof, true, &primaryError);
if (primaryError != 0)
ereport(ERROR,
(errcode_for_file_access(),
errmsg("file open error when opening file "
"'%s' for relation '%s': %s", filePathName, relname,
strerror(primaryError))));
*parquet_file_previous = mirroredOpen->primaryFile;
int64 fileSize = FileSeek(*parquet_file_previous, 0, SEEK_END);
if (fileSize < 0){
ereport(ERROR,
(errcode_for_file_access(),
errmsg("file seek error in file '%s' for relation "
"'%s'", filePathName, relname)));
}
if (logicalEof > fileSize) {
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR),
errmsg("logical eof exceeds file size in file '%s' for relation '%s'",
filePathName,
relname)));
}
/*read parquet footer, get metadata information before rowgroup metadata*/
metadataExist = readParquetFooter(*parquet_file_previous, parquetMetadata, protocol_read,
logicalEof, filePathName);
*previous_rowgroupcnt = (*parquetMetadata)->blockCount;
/*
* Open the file for writing.
*/
MirroredAppendOnly_OpenReadWrite(mirroredOpen, relFileNode, segmentFileNum,
relname, logicalEof, false, &primaryError);
if (primaryError != 0)
ereport(ERROR,
(errcode_for_file_access(),
errmsg("file open error when opening file '%s' "
"for relation '%s': %s", filePathName, relname,
strerror(primaryError))));
file = mirroredOpen->primaryFile;
seekResult = FileNonVirtualTell(file);
if (seekResult != logicalEof) {
/* previous transaction is aborted truncate file*/
if (FileTruncate(file, logicalEof)) {
MirroredAppendOnly_Close(mirroredOpen);
ereport(ERROR,
(errcode_for_file_access(),
errmsg("file truncate error in file '%s' for relation "
"'%s' to position " INT64_FORMAT ": %s",
filePathName, relname, logicalEof,
strerror(errno))));
}
}
*parquet_file = file;
/*if metadata not exist, should initialize the metadata, and write out file header*/
if (metadataExist == false) {
/* init parquet metadata information, init schema information using table attributes,
* and may get existing information from data file*/
initparquetMetadata(*parquetMetadata, tableAttrs, *parquet_file);
/*should judge whether file already exists, if a new file, should write header out*/
writeParquetHeader(*parquet_file, filePathName, fileLen, fileLen_uncompressed);
}
else {
if (!checkAndSyncMetadata(*parquetMetadata, tableAttrs))
{
ereport(ERROR,
(errcode(ERRCODE_GP_INTERNAL_ERROR),
errmsg("parquet storage write file's metadata incompatible "
"with table's schema for relation '%s'.",
relname)));
}
}
}
