zindex-py 0.0.4

Indexer for GZIP specially built for DLIO Profiler.

DISCLAIMER

This repo is a fork of the original repo located at https://github.com/mattgodbolt/zindex. We modify this repo for using it cohesively with DLIO Profiler https://github.com/hariharan-devarajan/dlio-profiler.

zindex creates and queries an index on a compressed, line-based text file in a time- and space-efficient way.

The itch I had

I have many multigigabyte text gzipped log files and I'd like to be able to find data in them by an index. There's a key on each line that a simple regex can pull out. However, to find a particular record requires zgrep, which takes ages as it has to seek through gigabytes of previous data to get to each record.

Enter zindex which builds an index and also stores decompression checkpoints along the way which allows lightning fast random access. Pulling out single lines by either line number of by an index entry is then almost instant, even for huge files. The indices themselves are small too, typically ~10% of the compressed file size for a simple unique numeric index.

Creating an index

zindex needs to be told what part of each line constitutes the index. This can be done by a regular expression, by field, or by piping each line through an external program.

By default zindex creates an index of file.gz.zindex when asked to index file.gz.

Example: create an index on lines matching a numeric regular expression. The capture group indicates the part that's to be indexed, and the options show each line has a unique, numeric index.

$ zindex file.gz --regex 'id:([0-9]+)' --numeric --unique

Example: create an index on the second field of a CSV file:

$ zindex file.gz --delimiter , --field 2

Example: create an index on a JSON field orderId.id in any of the items in the document root's actions array (requires jq). The jq query creates an array of all the orderId.ids, then joins them with a space to ensure each individual line piped to jq creates a single line of output, with multiple matches separated by spaces (which is the default separator).

$ zindex file.gz --pipe "jq --raw-output --unbuffered '[.actions[].orderId.id] | join(\" \")'"

Multiple indices, and configuration of the index creation by JSON configuration file are supported, see below.

Querying the index

The zq program is used to query an index. It's given the name of the compressed file and a list of queries. For example:

$ zq file.gz 1023 4443 554

It's also possible to output by line number, so to print lines 1 and 1000 from a file:

$ zq file.gz --line 1 1000

Building from source

zindex uses CMake for its basic building (though has a bootstrapping Makefile), and requires a C++11 compatible compiler (GCC 4.8 or above and clang 3.4 and above). It also requires zlib. With the relevant compiler available, building ought to be as simple as:

$ git clone https://github.com/mattgodbolt/zindex.git
$ cd zindex
$ make

Binaries are left in build/Release.

Additionally a static binary can be built if you're happy to dip your toe into CMake:

$ cd path/to/build/directory
$ cmake path/to/zindex/checkout/dir -DStatic:BOOL=On -DCMAKE_BUILD_TYPE=Release
$ make

Multiple indices

To support more than one index, or for easier configuration than all the command-line flags that might be needed, there is a JSON configuration format. Pass the --config <yourconfigfile>.json option and put something like this in the configuration file:

{ 
    "indexes": [
        {
            "type": "field",
            "delimiter": "\t",
            "fieldNum": 1
        },
        {
            "name": "secondary",
            "type": "field",
            "delimiter": "\t",
            "fieldNum": 2
        }
    ]
}

This creates two indices, one on the first field and one on the second field, as delimited by tabs. One can then specify which index to query with the -i <index> option of zq.

Issues and feature requests

See the issue tracker for TODOs and known bugs. Please raise bugs there, and feel free to submit suggestions there also.

Feel free to contact me if you prefer email over bug trackers.