share: Create a shared object

In SharedObject: Sharing R objects across multiple R processes without memory duplication

Description

This function will create a shared object in the shared memory for the object x. There is no duplication of the shared object when it is exported to the other processes.

Usage

share(x, ...)

## S4 method for signature 'ANY'
share(x, ..., copyOnWrite, sharedSubset, sharedCopy, mustWork)

Arguments

x	An R object that you want to shared, see details.
...	For generalization purpose.
copyOnWrite, sharedSubset, sharedCopy	The parameters controlling the behavior of the shared object, see details.
mustWork	Whether to throw an error if x is not a sharable object(e.g. Character). This parameter has no effect on the S4 object.

Details

The function returns a shared object corresponding to the argument x if it is sharable. There should be no different between x and the return value except that the latter one is shared. The attribute(s) of x will also be shared.

Supported types

For the basic R types, the function supports raw,logical ,integer, double, complex. character cannot be shared for it has a complicated data structure and closely relates to R's cache. For the containers, the function supports list, pairlist and environment. Note that sharing a container is equivalent sharing all elements in the container, the container itself will not be shared.

The function share is an S4 generic. The default share method works for most S4 objects. Therefore, there is no need to define a S4 share method for each S4 class unless the S4 class has a special implementation(e.g. on-disk data). The default method will share the object itself and all slots it contains. No error will be given if any of these objects are not sharable and they will be kept unchanged.

Behavior control

In the R level, the behaviors of an ALTREP object is exactly the same as an atomic object but the data of an ALTREP object is allocated in the shared memory space. Therefore an ALTREP object can be easily exported to the other R processes without dulplicating the data, which reduces the memory usage and the overhead of data transmission.

The behavior of a shared object can be controlled through three parameters: copyOnWrite, sharedSubset and sharedCopy.

copyOnWrite determines Whether a new R object need to be allocated when the shared object is changed. The default value is TRUE, but can be altered by passing an argument copyOnWrite = FALSE to the function.

Please note that the no-copy-on-write feature is not fully supported by R. When copyOnWrite is FALSE, a shared object might not behaves as user expects. Please refer to the example code to see the exceptions.

sharedSubset determines whether the subset of a shared object is still a shared object. The default value is FALSE, and can be changed by passing sharedSubset = TRUE to the function

At the time this documentation is being written, The shared subset feature will cause an unnecessary memory duplication in R studio. Therefore, for the performance consideration, it is better to turn the feature off when using R studio.

sharedCopy determines whether the object is still a shared object after a duplication. If copyOnWrite is FALSE, this feature is off since the duplication cannot be triggered. In current version (R 3.6), an object will be duplicated four times for creating a shared object and lead to a serious performance problem. Therefore, the default value is FALSE, user can alter it by passing sharedCopy = FALSE to the function.

Value

A shared object

Examples

## For vector
x <- runif(10)
so <- share(x)
x
so

## For matrix
x <- matrix(runif(10), 2, 5)
so <- share(x)
x
so

## For data frame
x <- as.data.frame(matrix(runif(10), 2, 5))
so <- share(x)
x
so

## export the object
library(parallel)
cl <- makeCluster(1)
clusterExport(cl, "so")
## check the exported object in the other process
clusterEvalQ(cl, so)

## close the connection
stopCluster(cl)

## Copy-on-write
## This is the default setting
x <- runif(10)
so1 <- share(x, copyOnWrite = TRUE)
so2 <- so1
so2[1] <- 10
## so1 is unchanged since copy-on-write feature is on.
so1
so2

## No-copy-on-write
so1 <- share(x, copyOnWrite = FALSE)
so2 <- so1
so2[1] <- 10
#so1 is changed
so1
so2

## Flaw of no-copy-on-write
## The following code changes the value of so1, highly unexpected! Please use with caution!
-so1
so1
## The reason is that the minus function trys to dulplicate so1 object,
## but the dulplicate function will return so1 itself, so the value in so1 also get changed.

SharedObject documentation built on Nov. 8, 2020, 8:17 p.m.