Learn R Programming

track (version 1.0-13)

track-package: Overview of track package

Description

The track package sets up a link between R objects in memory and files on disk so that objects are automatically saved to files when they are changed. R objects in files are read in on demand and do not consume memory prior to being referenced. The track package also tracks times when objects are created and modified, and caches some basic characteristics of objects to allow for fast summaries of objects. Each object is stored in a separate RData file using the standard format as used by save(), so that objects can be manually picked out of or added to the track database if needed. The track database is a directory usually named rdatadir that contains a RData file for each object and several housekeeping files that are either plain text or RData files. Tracking works by replacing a tracked variable by an activeBinding, which when accessed looks up information in an associated 'tracking environment' and reads or writes the corresponding RData file and/or gets or assigns the variable in the tracking environment. In the default mode of operation, R variables that are accessed are stored in memory for the duration of the top level task (i.e., in one expression evaluated from the prompt.) A callback that is called each time a top-level-task completes does three major things:
  • detects newly created or deleted variables, and adds or removes from the tracking database as appropriate, and
  • writes changed variables to the database, and
  • deletes cached objects from memory.
Tracking is not particularly suitable for storing objects that contain environments, because those environments and their contents will be fully written out in the saved file (in a live R session, environments are references, and there can be multiple references to one environment.) Functions are one of the most common objects that contain environments, which can contain data objects local to the function (e.g., see the examples in the R FAQ in the section "Lexical scoping" under "What are the differences between R and S?" http://cran.r-project.org/doc/FAQ/R-FAQ.html#Lexical-scoping). Additionally, the results of some modeling functions contain environments, e.g., lm holds several references to the environment that contains the data. When an lm object is save'ed, the environment containing the data, and all the other objects in that environment, can be saved in the same file. To work with large data objects and modeling functions, consider first creating a tracking database that contains the data objects. Then, in a different R session (which can be running at the same time), use track.attach to attach the db of data objects at pos=2 on the search list. When working in this way, the data objects will only be kept in memory when being used, and modeling functions that record environments in their results can be successful used (though beware of modeling functions that store large amounts of data in their results.) Alternatively, use modeling functions that do not store references to environments. The utility function show.envs from the track package will show what environments are referenced within an object (though it is not guaranteed to find them all.) The track package also provides a self-contained incremental history saving function that writes the most recent command to the file .Rincr_history at the end of each top-level task, along with a time stamp that does not appear in the interactive history. The standard history functionality (savehistory/loadhistory) in R writes the history only at the end of the session. Thus, if the R session terminates abnormally, history is lost.

Arguments

List of basic functions and common calling patterns

For straightforward use of the track package, only a single call to track.start() need be made to start automatically tracking the global environment. If it is desired to save untrackable variables at the end of the session, track.stop() should be called before calling save.image() or q('yes'), because track.stop() will ensure that tracked variables are saved to disk and then remove them from the global environment, leaving save.image() to save only the untracked or untrackable variables. The basic functions used in automatic tracking are as follows:
  • track.start(dir=...): start tracking the global environment, with files saved indir(the default isrdatadir).
  • track.summary(): print a summary of the basic characteristics of tracked variables: name, class, extent, and creation, modification and access times.
  • track.info(): print a summary of which tracking databases are currently active.
  • track.stop(pos=, all=): stop tracking. Any unsaved tracked variables are saved to disk. UnlesskeepVars=TRUEis supplied, all tracked variables become unavailable until tracking starts again.
  • track.attach(dir=..., pos=): attach an existing tracking database to the search list at the specified position. The default when attaching at positions other than 1 is to use readonly mode, but in non-readonly mode, changes to variables in the attached environment will be automatically saved to the database.
  • track.rescan(pos=): rescan a tracking directory that was attached bytrack.attach()at a position other than 1, and that is preferably readonly.
For the non-automatic mode, four other functions cover the majority of common usage:
  • track.start(dir=..., auto=TRUE/FALSE): start tracking the global environment, with files saved indir
  • track(x): start trackingx-xin the global environment is replaced by an active binding andxis saved in its corresponding file in the tracking directory and, if caching is on, in the tracking environment
  • track(x <- value): start trackingx
  • track(list=c('x', 'y')): start tracking specified variables
  • track(all=TRUE): start tracking all untracked variables in the global environment
  • untrack(x): stop tracking variablex- the R objectxis put back as an ordinary object in the global environment
  • untrack(all=TRUE): stop tracking all variables in the global environment (but tracking is still set up)
  • untrack(list=...): stop tracking specified variables
  • track.remove(x): completely remove all traces ofxfrom the global environment, tracking environment and tracking directory. Note that if variablexin the global environment is tracked,remove(x)will makexan "orphaned" variable:remove(x)will just remove the active binding from the global environment, and leavexin the tracked environment and on file, andxwill reappear after restarting tracking.

Complete list of functions and common calling patterns

The track package provides many additional functions for controlling how tracking is performed (e.g., whether or not tracked variables are cached in memory), examining the state of tracking (show which variables are tracked, untracked, orphaned, masked, etc.) and repairing tracking environments and databases that have become inconsistent or incomplete (this may result from resource limitiations, e.g., being unable to write a save file due to lack of disk space, or from manual tinkering, e.g., dropping a new save file into a tracking directory.) The functions that can be used to set up and take down tracking are:
  • track.start(dir=...): start tracking, using the supplied directory
  • track.stop(): stop tracking (any unsaved tracked variables are saved to disk and all tracked variables become unavailable until tracking starts again)
  • track.dir(): return the path of the tracking directory
Functions for tracking and stopping tracking variables:
  • track(x)track(var <- value)track(list=...)track(all=TRUE): start tracking variable(s)
  • track.load(file=...): load some objects from a RData file into the tracked environment
  • untrack(x, keep.in.db=FALSE)untrack(list=...)untrack(all=TRUE): stop tracking variable(s) - value is left in place, and optionally, it is also left in the the database
Functions for getting status of tracking and summaries of variables:
  • track.summary(): return a data frame containing a summary of the basic characteristics of tracked variables: name, class, extent, and creation, modification and access times.
  • track.status(): return a data frame containing information about the tracking status of variables: whether they are saved to disk or not, etc.
  • track.info(): return a data frame containing information about which tracking dbs are currently active.
  • env.is.tracked(): tell whether an environment is currently tracked
The remaining functions allow the user to more closely manage variable tracking, but are less likely to be of use to new users. Functions for getting status of tracking and summaries of variables:
  • tracked(): return the names of tracked variables
  • untracked(): return the names of untracked variables
  • untrackable(): return the names of variables that cannot be tracked
  • track.unsaved(): return the names of variables whose copy on file is out-of-date
  • track.orphaned(): return the names of once-tracked variables that have lost their active binding (should not happen)
  • track.masked(): return the names of once-tracked variables whose active binding has been overwritten by an ordinary variable (should not happen)
Functions for managing tracking and tracked variables:
  • track.options(): examine and set options to control tracking
  • track.remove(): completely remove all traces of a tracked variable
  • track.save(): write unsaved variables to disk
  • track.flush(): write unsaved variables to disk, and remove from memory
  • track.forget(): delete cached versions without saving to file (file version will be retrieved next time the variable is accessed)
  • track.rescan(): reload variable values from disk (can forget all cached vars, remove no-longer existing tracked vars)
  • track.load(): load variables from a saved RData file into the tracking session
  • track.copy()andtrack.move(): copy or move variables from one tracking db to another
  • track.rename()rename variables in a tracking db
Functions for recovering from errors:
  • track.rebuild(): rebuild tracking information from objects in memory or on disk
  • track.flush: write unsaved variables to disk, and remove from memory
Design and internals of tracking:

Details

There are four main reasons to use the track package:
  • conveniently handle many moderately-large objects that would collectively exhaust memory or be inconvenient to manage in files by manually usingsave(),load(), and/orsave.image().
  • have changed or newly created objects saved automatically at the end of each top-level command, which ensures objects are preserved in the event of accidental or abnormal termination of the R session, and which also makes startup and saving much faster when many large objects in the global environment must be loaded or saved.
  • keep track of creation and modification times on objects
  • get fast summaries of basic characteristics of objects - class, size, dimension, etc.
There is an option to control whether tracked objects are cached in memory as well as being stored on disk. By default, objects are cached in memory for the duration of a top-level task. To save time when working with collections of objects that will all fit in memory, turn on caching with and turn off cache-flushing track.options(cache=TRUE, cachePolicy="none"), or start tracking with track.start(..., cache=TRUE, cachePolicy="none"). A possible future improvement is to allow conditional and/or more intelligent caching of objects. Some data that would be needed for this is already collected in access counts and times that are recorded in the tracking summary. Here is a brief example of tracking some variables in the global environment: > library(track) > track.start() > x <- 123 # Variable 'x' is now tracked > y <- matrix(1:6, ncol=2) # 'y' is assigned & tracked > z1 <- list("a", "b", "c") > z2 <- Sys.time() > track.summary(size=F) # See a summary of tracked vars class mode extent length modified TA TW x numeric numeric [1] 1 2007-09-07 08:50:58 0 1 y matrix numeric [3x2] 6 2007-09-07 08:50:58 0 1 z1 list list [[3]] 3 2007-09-07 08:50:58 0 1 z2 POSIXt,POSIXct numeric [1] 1 2007-09-07 08:50:58 0 1 > # (TA="total accesses", TW="total writes") > ls(all=TRUE) [1] "x" "y" "z1" "z2" > track.stop(pos=1) # Stop tracking > ls(all=TRUE) character(0) > > # Restart using the tracking dir -- the variables reappear > track.start() # Start using the same tracking dir again ("rdatadir") > ls(all=TRUE) [1] "x" "y" "z1" "z2" > track.summary(size=F) class mode extent length modified TA TW x numeric numeric [1] 1 2007-09-07 08:50:58 0 1 y matrix numeric [3x2] 6 2007-09-07 08:50:58 0 1 z1 list list [[3]] 3 2007-09-07 08:50:58 0 1 z2 POSIXt,POSIXct numeric [1] 1 2007-09-07 08:50:58 0 1 > track.stop(pos=1) > > # the files in the tracking directory: > list.files("rdatadir", all=TRUE) [1] "." ".." [3] "filemap.txt" ".trackingSummary.rda" [5] "x.rda" "y.rda" [7] "z1.rda" "z2.rda" > There are several points to note:
  • The global environment is the default environment for tracking -- it is possible to track variables in other environments, but that environment must be supplied as an argument to the track functions.
  • By default, newly created or deleted variables are automatically added to or removed from the tracking database. This feature can be disabled by supplyingauto=FALSEtotrack.start(), or by callingtrack.auto(FALSE).
  • When tracking is stopped, all tracked variables are saved on disk and will be no longer accessible until tracking is started again.
  • The objects are stored each in their own file in the tracking dir, in the format used bysave()/load()(RData files).

References

Roger D. Peng. Interacting with data using the filehash package. R News, 6(4):19-24, October 2006. http://cran.r-project.org/doc/Rnews and http://sandybox.typepad.com/software David E. Brahm. Delayed data packages. R News, 2(3):11-12, December 2002. http://cran.r-project.org/doc/Rnews

See Also

Design of the track package. Potential future features of the track package. Documentation for save and load (in 'base' package). Documentation for makeActiveBinding and related functions (in 'base' package). Inspriation from the packages g.data and filehash. Description of the facility (addTaskCallback) for adding a callback function that is called at the end of each top-level task (each time R returns to the prompt after completing a command): http://developer.r-project.org/TaskHandlers.pdf.

Examples

Run this code
library(track)
# start tracking the global environment using directory 'rdatadir'
# inside dontrun to avoid creating/removing rdatadir
track.start()
a <- 1
b <- 2
ls()
track.status()
track.summary()
track.info()
track.stop()
# variables are now gone
ls()
# bring them back
track.start()
ls()
track.stop()

Run the code above in your browser using DataLab