// read or create index
private void readOrCreateIndex(CollectionManager.Force ff, Formatter f) throws IOException {
// force new index or test for new index needed
boolean force =
((ff == CollectionManager.Force.always)
|| (ff == CollectionManager.Force.test && needsUpdate()));
// otherwise, we're good as long as the index file exists
File idx = gc.getIndexFile();
if (force || !idx.exists() || !readIndex(idx.getPath())) {
logger.info("GribCollection {}: createIndex {}", gc.getName(), idx.getPath());
createIndex(idx, ff, f); // write out index
gc.rafLocation = idx.getPath();
gc.setRaf(new RandomAccessFile(idx.getPath(), "r"));
readIndex(gc.getRaf()); // read back in index
}
}
// LOOK what about extending an index ??
public boolean makeIndex(String filename, RandomAccessFile dataRaf, Formatter f)
throws IOException {
File idxFile = GribCollection.getIndexFile(filename + GBX9_IDX);
FileOutputStream fout = new FileOutputStream(idxFile);
RandomAccessFile raf = null;
try {
//// header message
fout.write(MAGIC_START.getBytes("UTF-8"));
NcStream.writeVInt(fout, version);
Map<Long, Integer> gdsMap = new HashMap<Long, Integer>();
gdsList = new ArrayList<Grib1SectionGridDefinition>();
records = new ArrayList<Grib1Record>(200);
Grib1IndexProto.Grib1Index.Builder rootBuilder = Grib1IndexProto.Grib1Index.newBuilder();
rootBuilder.setFilename(filename);
if (dataRaf == null) {
raf = new RandomAccessFile(filename, "r");
dataRaf = raf;
}
Grib1RecordScanner scan = new Grib1RecordScanner(dataRaf);
while (scan.hasNext()) {
Grib1Record r = scan.next();
if (r == null) break; // done
records.add(r);
Grib1SectionGridDefinition gds = r.getGDSsection();
Integer index = gdsMap.get(gds.calcCRC());
if (gds.getPredefinedGridDefinition()
>= 0) // skip predefined gds - they dont have raw bytes
index = 0;
else if (index == null) {
gdsList.add(gds);
index = gdsList.size() - 1;
gdsMap.put(gds.calcCRC(), index);
rootBuilder.addGdsList(makeGdsProto(gds));
}
rootBuilder.addRecords(makeRecordProto(r, index));
}
ucar.nc2.grib.grib1.Grib1IndexProto.Grib1Index index = rootBuilder.build();
byte[] b = index.toByteArray();
NcStream.writeVInt(fout, b.length); // message size
fout.write(b); // message - all in one gulp
f.format(" made gbx9 index for %s size=%d%n", filename, b.length);
return true;
} finally {
fout.close();
if (raf != null) raf.close();
}
}
private boolean needsUpdate(long collectionLastModified, Formatter f) throws IOException {
CollectionManager.ChangeChecker cc = Grib1Index.getChangeChecker();
for (CollectionManager dcm :
tpc.makePartitions()) { // LOOK not really right, since we dont know if these files are the
// same as in the index
File idxFile = GribCollection.getIndexFile(dcm);
if (!idxFile.exists()) return true;
if (collectionLastModified < idxFile.lastModified()) return true;
for (MFile mfile : dcm.getFiles()) {
if (cc.hasChangedSince(mfile, idxFile.lastModified())) return true;
}
}
return false;
}
protected GribCollection.VariableIndex readVariable(
GribCollectionProto.Variable pv, GribCollection.GroupHcs group) {
int discipline = pv.getDiscipline();
int category = pv.getCategory();
int param = pv.getParameter();
int levelType = pv.getLevelType();
int intvType = pv.getIntervalType();
String intvName = pv.getIntvName();
boolean isLayer = pv.getIsLayer();
int ensDerivedType = pv.getEnsDerivedType();
int probType = pv.getProbabilityType();
String probabilityName = pv.getProbabilityName();
int cdmHash = pv.getCdmHash();
long recordsPos = pv.getRecordsPos();
int recordsLen = pv.getRecordsLen();
int timeIdx = pv.getTimeIdx();
int vertIdx = pv.getVertIdx();
int ensIdx = pv.getEnsIdx();
int tableVersion = pv.getTableVersion();
return gc.makeVariableIndex(
group,
tableVersion,
discipline,
category,
param,
levelType,
isLayer,
intvType,
intvName,
ensDerivedType,
probType,
probabilityName,
cdmHash,
timeIdx,
vertIdx,
ensIdx,
recordsPos,
recordsLen);
}
/*
MAGIC_START
version
sizeRecords
VariableRecords (sizeRecords bytes)
sizeIndex
GribCollectionIndex (sizeIndex bytes)
*/
private boolean writeIndex(TimePartition.Partition canon, Formatter f) throws IOException {
File file = tp.getIndexFile();
if (file.exists()) {
if (!file.delete()) logger.error("Cant delete " + file.getPath());
}
RandomAccessFile raf = new RandomAccessFile(file.getPath(), "rw");
raf.order(RandomAccessFile.BIG_ENDIAN);
try {
//// header message
raf.write(MAGIC_START.getBytes("UTF-8"));
raf.writeInt(version);
raf.writeLong(0); // no record section
GribCollectionProto.GribCollectionIndex.Builder indexBuilder =
GribCollectionProto.GribCollectionIndex.newBuilder();
indexBuilder.setName(tp.getName());
GribCollection canonGc = canon.makeGribCollection(f);
for (GribCollection.GroupHcs g : canonGc.getGroups())
indexBuilder.addGroups(writeGroupProto(g));
indexBuilder.setCenter(canonGc.getCenter());
indexBuilder.setSubcenter(canonGc.getSubcenter());
indexBuilder.setMaster(canonGc.getMaster());
indexBuilder.setLocal(canonGc.getLocal());
for (TimePartition.Partition p : tp.getPartitions()) {
indexBuilder.addPartitions(writePartitionProto(p.getName(), (TimePartition.Partition) p));
}
GribCollectionProto.GribCollectionIndex index = indexBuilder.build();
byte[] b = index.toByteArray();
NcStream.writeVInt(raf, b.length); // message size
raf.write(b); // message - all in one gulp
f.format("GribCollectionTimePartitionedIndex= %d bytes%n", b.length);
} finally {
f.format("file size = %d bytes%n", raf.length());
raf.close();
}
return true;
}
// consistency check on variables : compare each variable to corresponding one in proto
// also set the groupno and partno for each partition
private boolean checkPartitions(TimePartition.Partition canon, Formatter f) throws IOException {
List<TimePartition.Partition> partitions = tp.getPartitions();
int npart = partitions.size();
boolean ok = true;
// for each group in canonical Partition
GribCollection canonGc = canon.makeGribCollection(f);
for (GribCollection.GroupHcs firstGroup : canonGc.getGroups()) {
String gname = firstGroup.getId();
if (trace) f.format(" Check Group %s%n", gname);
// hash proto variables for quick lookup
Map<Integer, GribCollection.VariableIndex> check =
new HashMap<Integer, GribCollection.VariableIndex>(firstGroup.varIndex.size());
List<GribCollection.VariableIndex> varIndexP =
new ArrayList<GribCollection.VariableIndex>(firstGroup.varIndex.size());
for (GribCollection.VariableIndex vi : firstGroup.varIndex) {
TimePartition.VariableIndexPartitioned vip = tp.makeVariableIndexPartitioned(vi, npart);
varIndexP.add(vip);
check.put(vi.cdmHash, vip); // replace with its evil twin
}
firstGroup.varIndex = varIndexP; // replace with its evil twin
// for each partition
for (int partno = 0; partno < npart; partno++) {
TimePartition.Partition tpp = partitions.get(partno);
if (trace) f.format(" Check Partition %s%n", tpp.getName());
// get corresponding group
GribCollection gc = tpp.makeGribCollection(f);
int groupIdx = gc.findGroupIdxById(firstGroup.getId());
if (groupIdx < 0) {
f.format(" Cant find group %s in partition %s%n", gname, tpp.getName());
ok = false;
continue;
}
GribCollection.GroupHcs group = gc.getGroup(groupIdx);
// for each variable in partition group
for (int varIdx = 0; varIdx < group.varIndex.size(); varIdx++) {
GribCollection.VariableIndex vi2 = group.varIndex.get(varIdx);
if (trace) f.format(" Check variable %s%n", vi2);
int flag = 0;
GribCollection.VariableIndex vi1 = check.get(vi2.cdmHash); // compare with proto variable
if (vi1 == null) {
f.format(
" WARN Cant find variable %s from %s in proto - ignoring that variable%n",
vi2, tpp.getName());
continue; // we can tolerate this
}
// compare vert coordinates
VertCoord vc1 = vi1.getVertCoord();
VertCoord vc2 = vi2.getVertCoord();
if ((vc1 == null) != (vc2 == null)) {
f.format(
" ERR Vert coordinates existence on variable %s in %s doesnt match%n",
vi2, tpp.getName());
ok = false;
} else if ((vc1 != null) && !vc1.equalsData(vc2)) {
f.format(
" WARN Vert coordinates values on variable %s in %s dont match%n",
vi2, tpp.getName());
f.format(" canon vc = %s%n", vc1);
f.format(" this vc = %s%n", vc2);
flag |= TimePartition.VERT_COORDS_DIFFER;
}
// compare ens coordinates
EnsCoord ec1 = vi1.getEnsCoord();
EnsCoord ec2 = vi2.getEnsCoord();
if ((ec1 == null) != (ec2 == null)) {
f.format(
" ERR Ensemble coordinates existence on variable %s in %s doesnt match%n",
vi2, tpp.getName());
ok = false;
} else if ((ec1 != null) && !ec1.equalsData(ec2)) {
f.format(
" WARN Ensemble coordinates values on variable %s in %s dont match%n",
vi2, tpp.getName());
f.format(" canon ec = %s%n", ec1);
f.format(" this ec = %s%n", ec2);
flag |= TimePartition.ENS_COORDS_DIFFER;
}
((TimePartition.VariableIndexPartitioned) vi1)
.setPartitionIndex(partno, groupIdx, varIdx, flag);
} // loop over variable
} // loop over partition
} // loop over group
if (ok) f.format(" Partition check: vert, ens coords OK%n");
return ok;
}
public boolean readIndex(String filename, long gribLastModified, CollectionManager.Force force)
throws IOException {
File idxFile = GribCollection.getIndexFile(filename + GBX9_IDX);
if (!idxFile.exists()) return false;
long idxModified = idxFile.lastModified();
if ((force != CollectionManager.Force.nocheck) && (idxModified < gribLastModified))
return false;
// force new index if file was updated
FileInputStream fin =
new FileInputStream(idxFile); // LOOK need DiskCache for non-writeable directories
try {
//// check header is ok
if (!NcStream.readAndTest(fin, MAGIC_START.getBytes())) {
log.info("Bad magic number of grib index, on file" + idxFile);
return false;
}
int v = NcStream.readVInt(fin);
if (v != version) {
if ((v == 0) || (v > version))
throw new IOException(
"Grib1Index found version " + v + ", want version " + version + " on " + filename);
if (log.isDebugEnabled())
log.debug(
"Grib1Index found version " + v + ", want version " + version + " on " + filename);
return false;
}
int size = NcStream.readVInt(fin);
if (size <= 0 || size > 100 * 1000 * 1000) { // try to catch garbage
log.warn("Grib1Index bad size = {} for {} ", size, filename);
return false;
}
byte[] m = new byte[size];
NcStream.readFully(fin, m);
Grib1IndexProto.Grib1Index proto = Grib1IndexProto.Grib1Index.parseFrom(m);
String fname = proto.getFilename();
if (debug) System.out.printf("%s for %s%n", fname, filename);
gdsList = new ArrayList<Grib1SectionGridDefinition>(proto.getGdsListCount());
for (Grib1IndexProto.Grib1GdsSection pgds : proto.getGdsListList()) {
Grib1SectionGridDefinition gds = readGds(pgds);
gdsList.add(gds);
}
if (debug) System.out.printf(" read %d gds%n", gdsList.size());
records = new ArrayList<Grib1Record>(proto.getRecordsCount());
for (Grib1IndexProto.Grib1Record precord : proto.getRecordsList()) {
records.add(readRecord(precord));
}
if (debug) System.out.printf(" read %d records%n", records.size());
} catch (java.lang.NegativeArraySizeException e) {
log.error("GribIndex failed on " + filename, e);
return false;
} catch (IOException e) {
log.error("GribIndex failed on " + filename, e);
return false;
} finally {
fin.close();
}
return true;
}
private void createIndex(
File indexFile, List<Group> groups, ArrayList<String> filenames, Formatter f)
throws IOException {
Grib2Record first = null; // take global metadata from here
if (indexFile.exists()) indexFile.delete(); // replace it
f.format(" createIndex for %s%n", indexFile.getPath());
RandomAccessFile raf = new RandomAccessFile(indexFile.getPath(), "rw");
raf.order(RandomAccessFile.BIG_ENDIAN);
try {
//// header message
raf.write(MAGIC_START.getBytes("UTF-8"));
raf.writeInt(version);
long lenPos = raf.getFilePointer();
raf.writeLong(0); // save space to write the length of the record section
long countBytes = 0;
int countRecords = 0;
for (Group g : groups) {
g.fileSet = new HashSet<Integer>();
for (Grib2Rectilyser.VariableBag vb : g.rect.getGribvars()) {
if (first == null) first = vb.first;
GribCollectionProto.VariableRecords vr = writeRecordsProto(vb, g.fileSet);
byte[] b = vr.toByteArray();
vb.pos = raf.getFilePointer();
vb.length = b.length;
raf.write(b);
countBytes += b.length;
countRecords += vb.recordMap.length;
}
}
long bytesPerRecord = countBytes / ((countRecords == 0) ? 1 : countRecords);
f.format(
" write RecordMaps: bytes = %d record = %d bytesPerRecord=%d%n",
countBytes, countRecords, bytesPerRecord);
if (first == null) {
logger.error("GribCollection {}: has no files\n{}", gc.getName(), f.toString());
throw new IllegalArgumentException("GribCollection " + gc.getName() + " has no files");
}
long pos = raf.getFilePointer();
raf.seek(lenPos);
raf.writeLong(countBytes);
raf.seek(pos); // back to the output.
GribCollectionProto.GribCollectionIndex.Builder indexBuilder =
GribCollectionProto.GribCollectionIndex.newBuilder();
indexBuilder.setName(gc.getName());
for (String fn : filenames) indexBuilder.addFiles(fn);
for (Group g : groups) indexBuilder.addGroups(writeGroupProto(g));
/* int count = 0;
for (DatasetCollectionManager dcm : collections) {
indexBuilder.addParams(makeParamProto(new Parameter("spec" + count, dcm.())));
count++;
} */
// what about just storing first ??
Grib2SectionIdentification ids = first.getId();
indexBuilder.setCenter(ids.getCenter_id());
indexBuilder.setSubcenter(ids.getSubcenter_id());
indexBuilder.setMaster(ids.getMaster_table_version());
indexBuilder.setLocal(ids.getLocal_table_version());
Grib2Pds pds = first.getPDS();
indexBuilder.setGenProcessType(pds.getGenProcessType());
indexBuilder.setGenProcessId(pds.getGenProcessId());
indexBuilder.setBackProcessId(pds.getBackProcessId());
GribCollectionProto.GribCollectionIndex index = indexBuilder.build();
byte[] b = index.toByteArray();
NcStream.writeVInt(raf, b.length); // message size
raf.write(b); // message - all in one gulp
f.format(" write GribCollectionIndex= %d bytes%n", b.length);
} finally {
f.format(" file size = %d bytes%n", raf.length());
raf.close();
if (raf != null) raf.close();
}
}
// read all records in all files,
// divide into groups based on GDS hash
// each group has an arraylist of all records that belong to it.
// for each group, run rectlizer to derive the coordinates and variables
public List<Group> makeAggregatedGroups(
ArrayList<String> filenames, CollectionManager.Force force, Formatter f) throws IOException {
Map<Integer, Group> gdsMap = new HashMap<Integer, Group>();
f.format("GribCollection %s: makeAggregatedGroups%n", gc.getName());
int total = 0;
int fileno = 0;
for (CollectionManager dcm : collections) {
// dcm.scanIfNeeded(); // LOOK ??
f.format(" dcm= %s%n", dcm);
Map<Integer, Integer> gdsConvert = (Map<Integer, Integer>) dcm.getAuxInfo("gdsHash");
for (MFile mfile : dcm.getFiles()) {
// f.format("%3d: %s%n", fileno, mfile.getPath());
filenames.add(mfile.getPath());
Grib2Index index = new Grib2Index();
try {
if (!index.readIndex(
mfile.getPath(),
mfile.getLastModified(),
force)) { // heres where the index date is checked against the data file
index.makeIndex(mfile.getPath(), f);
f.format(
" Index written: %s == %d records %n",
mfile.getName() + Grib2Index.IDX_EXT, index.getRecords().size());
} else if (debug) {
f.format(
" Index read: %s == %d records %n",
mfile.getName() + Grib2Index.IDX_EXT, index.getRecords().size());
}
} catch (IOException ioe) {
f.format(
"GribCollectionBuilder: reading/Creating gbx9 index failed err=%s%n skipping %s%n",
ioe.getMessage(), mfile.getPath() + Grib2Index.IDX_EXT);
continue;
}
for (Grib2Record gr : index.getRecords()) {
gr.setFile(fileno); // each record tracks which file it belongs to
int gdsHash =
gr.getGDSsection().getGDS().hashCode(); // use GDS hash code to group records
if (gdsConvert != null
&& gdsConvert.get(gdsHash)
!= null) { // allow external config to muck with gdsHash. Why? because of error in
// encoding
gdsHash = (Integer) gdsConvert.get(gdsHash); // and we need exact hash matching
}
Group g = gdsMap.get(gdsHash);
if (g == null) {
g = new Group(gr.getGDSsection(), gdsHash);
gdsMap.put(gdsHash, g);
}
g.records.add(gr);
total++;
}
fileno++;
}
}
f.format(" total grib records= %d%n", total);
Grib2Rectilyser.Counter c = new Grib2Rectilyser.Counter();
List<Group> result = new ArrayList<Group>(gdsMap.values());
for (Group g : result) {
g.rect = new Grib2Rectilyser(g.records, g.gdsHash);
f.format(" GDS hash %d == ", g.gdsHash);
g.rect.make(f, c);
}
f.format(
" Rectilyser: nvars=%d records unique=%d total=%d dups=%d (%f) %n",
c.vars, c.recordsUnique, c.records, c.dups, ((float) c.dups) / c.records);
return result;
}
public boolean readIndex(RandomAccessFile raf) {
gc.setRaf(raf); // LOOK leaving the raf open in the GribCollection
try {
raf.order(RandomAccessFile.BIG_ENDIAN);
raf.seek(0);
//// header message
if (!NcStream.readAndTest(raf, MAGIC_START.getBytes())) {
logger.error("GribCollection {}: invalid index", gc.getName());
return false;
}
int v = raf.readInt();
if (v != getVersion()) {
logger.warn(
"GribCollection {}: index found version={}, want version= {} on file {}",
new Object[] {gc.getName(), v, version, raf.getLocation()});
return false;
}
long skip = raf.readLong();
raf.skipBytes(skip);
int size = NcStream.readVInt(raf);
if ((size < 0) || (size > 100 * 1000 * 1000)) {
logger.warn("GribCollection {}: invalid index ", gc.getName());
return false;
}
byte[] m = new byte[size];
raf.readFully(m);
GribCollectionProto.GribCollectionIndex proto =
GribCollectionProto.GribCollectionIndex.parseFrom(m);
gc.center = proto.getCenter();
gc.subcenter = proto.getSubcenter();
gc.master = proto.getMaster();
gc.local = proto.getLocal();
gc.genProcessType = proto.getGenProcessType();
gc.genProcessId = proto.getGenProcessId();
gc.backProcessId = proto.getBackProcessId();
gc.local = proto.getLocal();
// gc.tables = Grib2Tables.factory(gc.center, gc.subcenter, gc.master, gc.local);
gc.filenames = new ArrayList<String>(proto.getFilesCount());
for (int i = 0; i < proto.getFilesCount(); i++) gc.filenames.add(proto.getFiles(i));
// error condition on a GribCollection Index
if ((proto.getFilesCount() == 0) && !(this instanceof TimePartitionBuilder)) {
logger.warn("GribCollection {}: has no files, force recreate ", gc.getName());
return false;
}
gc.groups = new ArrayList<GribCollection.GroupHcs>(proto.getGroupsCount());
for (int i = 0; i < proto.getGroupsCount(); i++)
gc.groups.add(readGroup(proto.getGroups(i), gc.makeGroup()));
Collections.sort(gc.groups);
gc.params = new ArrayList<Parameter>(proto.getParamsCount());
for (int i = 0; i < proto.getParamsCount(); i++) gc.params.add(readParam(proto.getParams(i)));
if (!readPartitions(proto)) {
logger.warn("TimePartition {}: has no partitions, force recreate ", gc.getName());
return false;
}
return true;
} catch (Throwable t) {
logger.error("Error reading index " + raf.getLocation(), t);
return false;
}
}
public boolean needsUpdate() {
File idx = gc.getIndexFile();
return !idx.exists() || needsUpdate(idx.lastModified());
}
// read all records in all files,
// divide into groups based on GDS hash
// each group has an arraylist of all records that belong to it.
// for each group, run rectlizer to derive the coordinates and variables
public List<Group> makeAggregatedGroups(
List<String> filenames, CollectionManager.Force force, Formatter f) throws IOException {
Map<Integer, Group> gdsMap = new HashMap<Integer, Group>();
boolean intvMerge = mergeIntvDefault;
f.format("GribCollection %s: makeAggregatedGroups%n", gc.getName());
int total = 0;
int fileno = 0;
for (CollectionManager dcm : collections) {
f.format(" dcm= %s%n", dcm);
FeatureCollectionConfig.GribConfig config =
(FeatureCollectionConfig.GribConfig)
dcm.getAuxInfo(FeatureCollectionConfig.AUX_GRIB_CONFIG);
Map<Integer, Integer> gdsConvert = (config != null) ? config.gdsHash : null;
FeatureCollectionConfig.GribIntvFilter intvMap = (config != null) ? config.intvFilter : null;
intvMerge =
(config == null) || (config.intvMerge == null) ? mergeIntvDefault : config.intvMerge;
for (MFile mfile : dcm.getFiles()) {
// f.format("%3d: %s%n", fileno, mfile.getPath());
filenames.add(mfile.getPath());
Grib2Index index = null;
try {
index =
(Grib2Index)
GribIndex.readOrCreateIndexFromSingleFile(
false, !isSingleFile, mfile, config, force, f);
} catch (IOException ioe) {
logger.warn(
"GribCollectionBuilder {}: reading/Creating gbx9 index failed err={}",
gc.getName(),
ioe.getMessage());
f.format(
"GribCollectionBuilder: reading/Creating gbx9 index failed err=%s%n skipping %s%n",
ioe.getMessage(), mfile.getPath() + GribIndex.IDX_EXT);
continue;
}
for (Grib2Record gr : index.getRecords()) {
if (this.tables == null) {
Grib2SectionIdentification ids =
gr.getId(); // so all records must use the same table (!)
this.tables =
Grib2Customizer.factory(
ids.getCenter_id(),
ids.getSubcenter_id(),
ids.getMaster_table_version(),
ids.getLocal_table_version());
if (config != null)
tables.setTimeUnitConverter(
config
.getTimeUnitConverter()); // LOOK doesnt really work with multiple collections
}
if (intvMap != null && filterTinv(gr, intvMap, f)) continue; // skip
gr.setFile(fileno); // each record tracks which file it belongs to
int gdsHash =
gr.getGDSsection().getGDS().hashCode(); // use GDS hash code to group records
if (gdsConvert != null
&& gdsConvert.get(gdsHash)
!= null) // allow external config to muck with gdsHash. Why? because of error in
// encoding
gdsHash = (Integer) gdsConvert.get(gdsHash); // and we need exact hash matching
Group g = gdsMap.get(gdsHash);
if (g == null) {
g = new Group(gr.getGDSsection(), gdsHash);
gdsMap.put(gdsHash, g);
}
g.records.add(gr);
total++;
}
fileno++;
}
}
f.format(" total grib records= %d%n", total);
Grib2Rectilyser.Counter c = new Grib2Rectilyser.Counter(); // debugging
List<Group> result = new ArrayList<Group>(gdsMap.values());
for (Group g : result) {
g.rect = new Grib2Rectilyser(tables, g.records, g.gdsHash, intvMerge);
f.format(" GDS hash %d == ", g.gdsHash);
g.rect.make(f, c, filenames);
}
f.format(
" Rectilyser: nvars=%d records unique=%d total=%d dups=%d (%f) %n",
c.vars, c.recordsUnique, c.records, c.dups, ((float) c.dups) / c.records);
return result;
}