INCR

Syntax

  1. INCR key

Time complexity: O(1)

Increments the number stored at key by one. If the key does not exist, it is set to 0 before performing the operation. An error is returned if the key contains a value of the wrong type or contains a string that can not be represented as integer. This operation is limited to 64 bit signed integers.

Note: this is a string operation because Dragonfly does not have a dedicated integer type. The string stored at the key is interpreted as a base-10 64 bit signed integer to execute the operation.

Dragonfly stores integers in their integer representation, so for string values that actually hold an integer, there is no overhead for storing the string representation of the integer.

Return

Integer reply: the value of key after the increment

Examples

  1. dragonfly> SET mykey "10"
  2. "OK"
  3. dragonfly> INCR mykey
  4. (integer) 11
  5. dragonfly> GET mykey
  6. "11"

Pattern: Counter

The counter pattern is the most obvious thing you can do with atomic increment operations. The idea is simply send an INCR command to Dragonfly every time an operation occurs. For instance in a web application we may want to know how many page views this user did every day of the year.

To do so the web application may simply increment a key every time the user performs a page view, creating the key name concatenating the User ID and a string representing the current date.

This simple pattern can be extended in many ways:

  • It is possible to use INCR and EXPIRE together at every page view to have a counter counting only the latest N page views separated by less than the specified amount of seconds.
  • A client may use GETSET in order to atomically get the current counter value and reset it to zero.
  • By using other atomic increment/decrement commands like DECR or INCRBY, it is possible to handle values that may get bigger or smaller depending on the operations performed by the user. Imagine for instance the score of different users in an online game.

Pattern: Rate limiter

The rate limiter pattern is a special counter that is used to limit the rate at which an operation can be performed. The classical materialization of this pattern involves limiting the number of requests that can be performed against a public API.

We provide two implementations of this pattern using INCR, where we assume that the problem to solve is limiting the number of API calls to a maximum of ten requests per second per IP address.

Pattern: Rate limiter 1

The more simple and direct implementation of this pattern is the following:

  1. FUNCTION LIMIT_API_CALL(ip)
  2. ts = CURRENT_UNIX_TIME()
  3. keyname = ip+":"+ts
  4. MULTI
  5.     INCR(keyname)
  6.     EXPIRE(keyname,10)
  7. EXEC
  8. current = RESPONSE_OF_INCR_WITHIN_MULTI
  9. IF current > 10 THEN
  10.     ERROR "too many requests per second"
  11. ELSE
  12.     PERFORM_API_CALL()
  13. END

Basically we have a counter for every IP, for every different second. But this counters are always incremented setting an expire of 10 seconds so that they’ll be removed by Redis automatically when the current second is a different one.

Note the used of MULTI and EXEC in order to make sure that we’ll both increment and set the expire at every API call.

Pattern: Rate limiter 2

An alternative implementation uses a single counter, but is a bit more complex to get it right without race conditions. We’ll examine different variants.

  1. FUNCTION LIMIT_API_CALL(ip):
  2. current = GET(ip)
  3. IF current != NULL AND current > 10 THEN
  4.     ERROR "too many requests per second"
  5. ELSE
  6.     value = INCR(ip)
  7.     IF value == 1 THEN
  8.         EXPIRE(ip,1)
  9.     END
  10.     PERFORM_API_CALL()
  11. END

The counter is created in a way that it only will survive one second, starting from the first request performed in the current second. If there are more than 10 requests in the same second the counter will reach a value greater than 10, otherwise it will expire and start again from 0.

In the above code there is a race condition. If for some reason the client performs the INCR command but does not perform the EXPIRE the key will be leaked until we’ll see the same IP address again.

This can be fixed easily turning the INCR with optional EXPIRE into a Lua script that is send using the EVAL command (only available since Redis version 2.6).

  1. local current
  2. current = redis.call("incr",KEYS[1])
  3. if current == 1 then
  4.     redis.call("expire",KEYS[1],1)
  5. end

There is a different way to fix this issue without using scripting, by using Redis lists instead of counters. The implementation is more complex and uses more advanced features but has the advantage of remembering the IP addresses of the clients currently performing an API call, that may be useful or not depending on the application.

  1. FUNCTION LIMIT_API_CALL(ip)
  2. current = LLEN(ip)
  3. IF current > 10 THEN
  4.     ERROR "too many requests per second"
  5. ELSE
  6.     IF EXISTS(ip) == FALSE
  7.         MULTI
  8.             RPUSH(ip,ip)
  9.             EXPIRE(ip,1)
  10.         EXEC
  11.     ELSE
  12.         RPUSHX(ip,ip)
  13.     END
  14.     PERFORM_API_CALL()
  15. END

The RPUSHX command only pushes the element if the key already exists.

Note that we have a race here, but it is not a problem: EXISTS may return false but the key may be created by another client before we create it inside the MULTI / EXEC block. However this race will just miss an API call under rare conditions, so the rate limiting will still work correctly.