a Temporary solution for xdb size exceeds the INT.max which will cause nagetive array index
This commit is contained in:
parent
bff6b5661c
commit
1d32d5e6aa
|
|
@ -95,6 +95,7 @@ public class SearchTest {
|
|||
long cost = TimeUnit.NANOSECONDS.toMicros((long) (System.nanoTime() - sTime));
|
||||
System.out.printf("{region: %s, ioCount: %d, took: %d μs}\n", region, searcher.getIOCount(), cost);
|
||||
} catch (Exception e) {
|
||||
e.printStackTrace();
|
||||
System.out.printf("{err: %s, ioCount: %d}\n", e, searcher.getIOCount());
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -81,33 +81,33 @@ public class Searcher {
|
|||
this.ioCount = 0;
|
||||
|
||||
// locate the segment index block based on the vector index
|
||||
int sPtr = 0, ePtr = 0;
|
||||
long sPtr = 0, ePtr = 0;
|
||||
int il0 = (int) ((ip >> 24) & 0xFF);
|
||||
int il1 = (int) ((ip >> 16) & 0xFF);
|
||||
int idx = il0 * VectorIndexCols * VectorIndexSize + il1 * VectorIndexSize;
|
||||
// System.out.printf("il0: %d, il1: %d, idx: %d\n", il0, il1, idx);
|
||||
if (vectorIndex != null) {
|
||||
sPtr = getInt(vectorIndex, idx);
|
||||
ePtr = getInt(vectorIndex, idx + 4);
|
||||
sPtr = getIntLong(vectorIndex, idx);
|
||||
ePtr = getIntLong(vectorIndex, idx + 4);
|
||||
} else if (contentBuff != null) {
|
||||
sPtr = getInt(contentBuff, HeaderInfoLength + idx);
|
||||
ePtr = getInt(contentBuff, HeaderInfoLength + idx + 4);
|
||||
sPtr = getIntLong(contentBuff, HeaderInfoLength + idx);
|
||||
ePtr = getIntLong(contentBuff, HeaderInfoLength + idx + 4);
|
||||
} else {
|
||||
final byte[] buff = new byte[VectorIndexSize];
|
||||
read(HeaderInfoLength + idx, buff);
|
||||
sPtr = getInt(buff, 0);
|
||||
ePtr = getInt(buff, 4);
|
||||
sPtr = getIntLong(buff, 0);
|
||||
ePtr = getIntLong(buff, 4);
|
||||
}
|
||||
|
||||
// System.out.printf("sPtr: %d, ePtr: %d\n", sPtr, ePtr);
|
||||
|
||||
// binary search the segment index block to get the region info
|
||||
final byte[] buff = new byte[SegmentIndexSize];
|
||||
int dataLen = -1, dataPtr = -1;
|
||||
int l = 0, h = (ePtr - sPtr) / SegmentIndexSize;
|
||||
int dataLen = -1;
|
||||
long dataPtr = -1, l = 0, h = (ePtr - sPtr) / SegmentIndexSize;
|
||||
while (l <= h) {
|
||||
int m = (l + h) >> 1;
|
||||
int p = sPtr + m * SegmentIndexSize;
|
||||
long m = (l + h) >> 1;
|
||||
long p = sPtr + m * SegmentIndexSize;
|
||||
|
||||
// read the segment index
|
||||
read(p, buff);
|
||||
|
|
@ -120,14 +120,14 @@ public class Searcher {
|
|||
l = m + 1;
|
||||
} else {
|
||||
dataLen = getInt2(buff, 8);
|
||||
dataPtr = getInt(buff, 10);
|
||||
dataPtr = getIntLong(buff, 10);
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// empty match interception
|
||||
// System.out.printf("dataLen: %d, dataPtr: %d\n", dataLen, dataPtr);
|
||||
System.out.printf("dataLen: %d, dataPtr: %d\n", dataLen, dataPtr);
|
||||
if (dataPtr < 0) {
|
||||
return null;
|
||||
}
|
||||
|
|
@ -138,11 +138,18 @@ public class Searcher {
|
|||
return new String(regionBuff, "utf-8");
|
||||
}
|
||||
|
||||
protected void read(int offset, byte[] buffer) throws IOException {
|
||||
protected void read(long offset, byte[] buffer) throws IOException {
|
||||
// check the in-memory buffer first
|
||||
if (contentBuff != null) {
|
||||
// @TODO: reduce data copying, directly decode the data ?
|
||||
System.arraycopy(contentBuff, offset, buffer, 0, buffer.length);
|
||||
// @TODO: added by Leon at 2025/06/10, when offset is negative and the content byte is not going to work.
|
||||
// we need a better solution for the content buffer which is greater than (2^31 - 1 << 2)
|
||||
int int_idx = (int) offset;
|
||||
if (int_idx < 0) {
|
||||
throw new IOException("No content buffer policy for NOW since the xdb is too large, use file or vectorIndex instead");
|
||||
}
|
||||
|
||||
System.arraycopy(contentBuff, int_idx, buffer, 0, buffer.length);
|
||||
return;
|
||||
}
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue