Splitting Data for Distributed Processing¶
Before you read this section, make sure that you understand the concepts described in Distributed Computing.
Pachyderm enables you to parallelize computations over data as long as that data can be split up into multiple datums. However, in many cases, you might have a dataset that you want or need to commit into Pachyderm as a single file rather than a bunch of smaller files that are easily mapped to datums, such as one file per record. For such cases, Pachyderm provides an easy way to prepare your dataset for subsequent distributed computing by splitting it upon uploading to a Pachyderm repository.
In this example, you have a dataset that consists of information about your users and a repository called
user. This data is in
CSV format in a single file called
user_data.csv with one record per line:
head user_data.csv 1,firstname.lastname@example.org,22.214.171.124 2,email@example.com,126.96.36.199 3,firstname.lastname@example.org,188.8.131.52 4,email@example.com,184.108.40.206 5,firstname.lastname@example.org,220.127.116.11 6,email@example.com,18.104.22.168 7,firstname.lastname@example.org,22.214.171.124 8,email@example.com,126.96.36.199 9,firstname.lastname@example.org,188.8.131.52 10,email@example.com,184.108.40.206
If you put this data into Pachyderm as a single file, Pachyderm processes them a single datum. It cannot process each of these user records in parallel as separate
datums. Potentially, you can manually separate these user records into standalone files before you commit them into the
users repository or through a pipeline stage dedicated to this splitting task. However, Pachyderm provides an optimized way of completing this task.
put file API includes an option for splitting the file into separate datums automatically. You can use the
--split flag with the
put file command.
To complete this example, follow the steps below:
usersrepository by running:
pachctl create repo users
Create a file called
user_data.csvwith the contents listed above.
user_data.csvfile into Pachyderm and automatically split it into separate datums for each line:
pachctl put file users@master -f user_data.csv --split line --target-file-datums 1
--split lineargument specifies that Pachyderm splits this file into lines, and the
--target-file-datums 1argument specifies that each resulting file must include at most one datum or one line.
View the list of files in the master branch of the
pachctl list file users@master
NAME TYPE SIZE user_data.csv dir 5.346 KiB
If you run
pachctl list filecommand for the master branch in the
usersrepository, Pachyderm still shows the
user_data.csventity to you as one entity in the repo However, this entity is now a directory that contains all of the split records.
To view the detailed information about the
user_data.csvfile, run the command with the file name specified after a colon:
pachctl list file users@master:user_data.csv
NAME TYPE SIZE user_data.csv/0000000000000000 file 43 B user_data.csv/0000000000000001 file 39 B user_data.csv/0000000000000002 file 37 B user_data.csv/0000000000000003 file 34 B user_data.csv/0000000000000004 file 35 B user_data.csv/0000000000000005 file 41 B user_data.csv/0000000000000006 file 32 B etc...
Then, a pipeline that takes the repo
usersas input with a glob pattern of
/user_data.csv/*processes each user record, such as each line in the CSV file in parallel.
JSON and Text File Splitting Examples¶
Pachyderm supports this type of splitting for lines or JSON blobs as well. See the examples below.
- Split a
jsonblobs by putting each
jsonblob into a separate file.
pachctl put file users@master -f user_data.json --split json --target-file-datums 1
- Split a
jsonblobs by putting three
jsonblobs into each split file.
pachctl put file users@master -f user_data.json --split json --target-file-datums 3
- Split a file on lines by putting each 100-bytes chunk into the split files.
pachctl put file users@master -f user_data.txt --split line --target-file-bytes 100
Specifying a Header¶
If your data has a common header, you can specify it manually by using
pachctl put file with the
--header-records flag. You can use this functionality with JSON and CSV data.
To specify a header, complete the following steps:
Create a new or use an existing data file. For example, the
user_data.csvfrom the section above with the following header:
Create a new repository or use an existing one:
pachctl create repo users
Put your file into the repository by separating the header from other lines:
pachctl put file users@master -f user_data.csv --split=csv --header-records=1 --target-file-datums=1
Verify that the file was added and split:
pachctl list file users@master:/user_data.csv
NAME TYPE SIZE /user_data.csv/0000000000000000 file 70B /user_data.csv/0000000000000001 file 66B /user_data.csv/0000000000000002 file 64B /user_data.csv/0000000000000003 file 61B /user_data.csv/0000000000000004 file 62B /user_data.csv/0000000000000005 file 68B /user_data.csv/0000000000000006 file 59B /user_data.csv/0000000000000007 file 59B /user_data.csv/0000000000000008 file 71B /user_data.csv/0000000000000009 file 65B
Get the first file from the repository:
pachctl get file users@master:/user_data.csv/0000000000000000
Get all files:
pachctl get file users@master:/user_data.csv/*
NUMBER,EMAIL,IP_ADDRESS 1,firstname.lastname@example.org,220.127.116.11 2,email@example.com,18.104.22.168 3,firstname.lastname@example.org,22.214.171.124 4,email@example.com,126.96.36.199 5,firstname.lastname@example.org,188.8.131.52 6,email@example.com,184.108.40.206 7,firstname.lastname@example.org,220.127.116.11 8,email@example.com,18.104.22.168 9,firstname.lastname@example.org,22.214.171.124 10,email@example.com,126.96.36.199
For more information, type
pachctl put file --help.
Ingesting PostgresSQL data¶
Pachyderm supports direct data ingestion from PostgreSQL. You need first extract your database into a script file by using
pg_dump and then add the data from the file into Pachyderm by running the
pachctl put file with the
When you use
pachctl put file --split sql ..., Pachyderm splits your
pgdump file into three parts - the header, rows, and the footer. The header contains all the SQL statements in the
pgdump file that set up the schema and tables. The rows are split into individual files, or if you specify the
--target-file-bytes, multiple rows per file. The footer contains the remaining SQL statements for setting up the tables.
The header and footer are stored in the directory that contains the rows. If you request a
get file on that directory, you get just the header and footer. If you request an individual file, you see the header, the row or rows, and the footer. If you request all the files with a glob pattern, for example,
/directoryname/*, you receive the header, all the rows, and the footer recreating the full
pgdump. Therefore, you can construct full or partial
pgdump files so that you can load full or partial datasets.
To put your PostgreSQL data into Pachyderm, complete the following steps:
pg_dump -t users -f users.pgdump
-- -- PostgreSQL database dump -- -- Dumped from database version 9.5.12 -- Dumped by pg_dump version 9.5.12 SET statement_timeout = 0; SET lock_timeout = 0; SET client_encoding = 'UTF8'; SET standard_conforming_strings = on; SELECT pg_catalog.set_config('search_path', '', false); SET check_function_bodies = false; SET client_min_messages = warning; SET row_security = off; SET default_tablespace = ''; SET default_with_oids = false; -- -- Name: users; Type: TABLE; Schema: public; Owner: postgres -- CREATE TABLE public.users ( id integer NOT NULL, name text NOT NULL, saying text NOT NULL ); ALTER TABLE public.users OWNER TO postgres; -- -- Data for Name: users; Type: TABLE DATA; Schema: public; Owner: postgres -- COPY public.users (id, name, saying) FROM stdin; 0 wile E Coyote ... 1 road runner \\. \. -- -- PostgreSQL database dump complete --
Ingest the SQL data by using the
pachctl put filecommand with the
pachctl put file data@master -f users.pgdump --split sql
pachctl put file data@master:users --split sql -f users.pgdump
View the information about your repository:
pachctl list file data@master
NAME TYPE SIZE users dir 914B
users.pgdumpfile is added to the master branch in the
View the information about the
pachctl list file data@master:users
NAME TYPE SIZE /users/0000000000000000 file 20B /users/0000000000000001 file 18B
In your pipeline, where you have started and forked PostgreSQL, you can load the data by running the following or a similar script:
cat /pfs/data/users/* | sudo -u postgres psql
By using the glob pattern
/*, this code loads each raw PostgreSQL chunk into your PostgreSQL instance for processing by your pipeline.
For this use case, you might want to use
--target-file-bytes because these commands enable your queries to run against many rows at a time.