Migrate Redis data on Amazon ElastiCache

migrate-redis.py

"""
Copies all keys from the source Redis host to the destination Redis host.
Useful to migrate Redis instances where commands like SLAVEOF and MIGRATE are
restricted (e.g. on Amazon ElastiCache).

The script scans through the keyspace of the given database number and uses
a pipeline of DUMP and RESTORE commands to migrate the keys.

Requires Redis 2.8.0 or higher.

Python requirements:
click==4.0
progressbar==2.3
redis==2.10.3
"""

import click
from progressbar import ProgressBar
from progressbar.widgets import Percentage, Bar, ETA
import redis
from redis.exceptions import ResponseError

@click.command()
@click.argument('srchost')
@click.argument('dsthost')
@click.option('--db', default=0, help='Redis db number, default 0')
@click.option('--flush', default=False, is_flag=True, help='Delete all keys from destination before migrating')
def migrate(srchost, dsthost, db, flush):
    if srchost == dsthost:
        print 'Source and destination must be different.'
        return

    source = redis.Redis(srchost, db=db)
    dest = redis.Redis(dsthost, db=db)

    if flush:
        dest.flushdb()

    size = source.dbsize()

    if size == 0:
        print 'No keys found.'
        return

    progress_widgets = ['%d keys: ' % size, Percentage(), ' ', Bar(), ' ', ETA()]
    pbar = ProgressBar(widgets=progress_widgets, maxval=size).start()

    COUNT = 2000 # scan size

    cnt = 0
    non_existing = 0
    already_existing = 0
    cursor = 0

    while True:
        cursor, keys = source.scan(cursor, count=COUNT)
        pipeline = source.pipeline()
        for key in keys:
            pipeline.pttl(key)
            pipeline.dump(key)
        result = pipeline.execute()

        pipeline = dest.pipeline()

        for key, ttl, data in zip(keys, result[::2], result[1::2]):
            if ttl is None:
                ttl = 0
            if data != None:
                pipeline.restore(key, ttl, data)
            else:
                non_existing += 1

        results = pipeline.execute(False)
        for key, result in zip(keys, results):
            if result != 'OK':
                e = result
                if hasattr(e, 'message') and (e.message == 'BUSYKEY Target key name already exists.' or e.message == 'Target key name is busy.'):
                    already_existing += 1
                else:
                    print 'Key failed:', key, `data`, `result`
                    raise e

        if cursor == 0:
            break

        cnt += len(keys)
        pbar.update(min(size, cnt))

    pbar.finish()
    print 'Keys disappeared on source during scan:', non_existing
    print 'Keys already existing on destination:', already_existing

if __name__ == '__main__':
    migrate()

You saved us a hassle and a half as MIGRATE is disabled on AWS Elasticache! Thank you so much

I used this script successfully to migrate data on multiple AWS environments, but one instance (with the most data) is failing due to

  File "/usr/local/lib/python2.7/site-packages/redis/connection.py", line 208, in read
    return data[:-2]
MemoryError

Any clues on what could be tweaked to migrate this redis instance?

Thank you @thomasst!

Just in case anybody stumbles upon this gist, I forked @kitwalker12's fork (thanks too!), updated it to support Python 3 and added a couple of options to handle SSL connections and basic filtering on the names of the keys to migrate. The result can be found here :)

👍

is there any way we can override the values of keys when the value of key in the source is changed and running the script again. Currently if i run the script second time after the value of key is changed in the source, that change is not getting reflected in the second host(destination)

Hello, you can use restore(name, ttl, value, replace=True) for overriding the keys everytime you copy.

I'm getting the error redis.exceptions.ResponseError: DUMP payload version or checksum are wrong, any idea what that means?

Thanks a lot! 👍

Thanks a ton. You made migrating on Elasticache easier. I tried a lot of other ways but this was the fastest option.

Super faster