AMC-team created HADOOP-19673:
---------------------------------
Summary: BloomMapFile: invalid io.mapfile.bloom.error.rate (≤0 or
≥1) causes NaN/zero vector size and writer construction failure
Key: HADOOP-19673
URL: https://issues.apache.org/jira/browse/HADOOP-19673
Project: Hadoop Common
Issue Type: Bug
Components: common, io
Affects Versions: 2.8.5
Reporter: AMC-team
{{BloomMapFile.Writer#initBloomFilter(Configuration)}} computes the Bloom
filter vector size as:
{code:java}
int numKeys = conf.getInt(IO_MAPFILE_BLOOM_SIZE_KEY,
IO_MAPFILE_BLOOM_SIZE_DEFAULT);
float errorRate = conf.getFloat(IO_MAPFILE_BLOOM_ERROR_RATE_KEY,
IO_MAPFILE_BLOOM_ERROR_RATE_DEFAULT);// vectorSize = ceil( -k * n / ln(1 -
p^(1/k)) )
int vectorSize = (int) Math.ceil(
(double)(-HASH_COUNT * numKeys) /
Math.log(1.0 - Math.pow(errorRate, 1.0 / HASH_COUNT))
); {code}
When {{io.mapfile.bloom.error.rate}} is *≤ 0* or {*}≥ 1{*}:
* {{Math.pow(errorRate, 1/k)}} produces *NaN* (negative base with non-integer
exponent) or an invalid value;
* {{Math.log(1 - NaN)}} becomes {*}NaN{*};
* {{Math.ceil(NaN)}} cast to {{int}} yields {*}0{*}, so {{{}vectorSize ==
0{}}};
* constructing {{DynamicBloomFilter}} subsequently fails, and
{{BloomMapFile.Writer}} construction fails (observed as assertion failure in
tests).
The code misses input validation for {{io.mapfile.bloom.error.rate}} which
should be strictly within {{{}(0, 1){}}}. With invalid values, the math
silently degrades to NaN/0 and fails at runtime.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: common-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-dev-h...@hadoop.apache.org
