orderly bundles

If you have an orderly report that takes a very long time, or needs to run in parallel, you might need to send it to run on another computer. There are a number of ways of achieving this - the simplest might be to clone the source tree to another computer, run the reports there and use one of a number of possible approaches to sync the outputs between computers. However, there will be cases where that is not ideal, and you want to move around much less data around.

This vignette describes a way of parcelling together all dependencies of an orderly report into a zip file (a “bundle”) that can be distributed to another machine (e.g., via scp, rsync or a shared file system), run there, and returned. It does not provide a transparent approach to using high-performance computing with orderly as we feel that the specific circumstances are too varied to support this directly.

Overview

In order to use orderly bundles we make some assumptions and conventions.

First, we assume that you will be running your exported report on another machine (otherwise you would have access to the orderly tree) and that your report takes really quite a long time.

Second, we assume you have your own way of getting the bundled reports to your other machine and the completed bundles back again — we expect that the details here will be specific to your needs and situation and that the overhead of doing this will be trivial compared with the cost of running the report.

Third, we assume that you will deal with all issues around queuing, locking and fault tolerance. From orderly’s point of view work will be exported from orderly and at this point you’re in control - we expect it to come back computed at some point, though we do not enforce that.

Fourth, that the machine running the report can be trusted to actually run the report - if you have set up an orderly server that is safe from interactively run reports, don’t allow importing from anyone’s laptops if you want to preserve this.

Fifth, that you trust your remote machine with your data, and that the remote machine trusts your orderly archive enough to run arbitrary code on it.

The process

We will use the orderly demo example, and pack up the use_dependency report.

path <- orderly::orderly_example("demo")

The use_dependency report has a dependency, which we run

id <- orderly::orderly_run("other", parameters = list(nmin = 0),
                           echo = FALSE, root = path)
## [ name       ]  other
## [ id         ]  20241018-053120-f01c0175
## [ sources    ]  functions.R
## [ parameter  ]  nmin: 0
## [ start      ]  2024-10-18 05:31:20.958863
## [ data       ]  source => extract: 20 x 2
## [ parameter  ]  nmin: 0
## [ end        ]  2024-10-18 05:31:21.032694
## [ elapsed    ]  Ran report in 0.0738306 secs
## [ artefact   ]  summary.csv: 3fac8347e152c84c96e6676413c718b7
## [ ...        ]  graph.png: 67b3e662440b3978ee78b9d4cc159884
orderly::orderly_commit(id, root = path)
## [ commit     ]  other/20241018-053120-f01c0175
## [ copy       ]
## [ import     ]  other:20241018-053120-f01c0175
## [ success    ]  :)
## [1] "/tmp/RtmpTaFezJ/filee0f695f49e0/archive/other/20241018-053120-f01c0175"

We need a place that we’ll put the bundles:

path_bundles <- tempfile()

Now, we can pack up use_dependency to run

bundle <- orderly::orderly_bundle_pack(path_bundles, "use_dependency",
                                       root = path)
## [ name       ]  use_dependency
## [ id         ]  20241018-053121-57723cc2
## [ depends    ]  other@20241018-053120-f01c0175:summary.csv -> incoming.csv
## [ start      ]  2024-10-18 05:31:21.344649
## [ bundle pack ]  20241018-053121-57723cc2
bundle
## $id
## [1] "20241018-053121-57723cc2"
## 
## $path
## [1] "/tmp/RtmpTaFezJ/filee0f13df5ad7/20241018-053121-57723cc2.zip"

orderly_bundle_pack has created a zip file. The format of this file is internal to orderly (it will likely change and will at some point become resistant to tampering), but contains:

zip::zip_list(bundle$path)
##                                      filename compressed_size uncompressed_size
## 1                   20241018-053121-57723cc2/               0                 0
## 2              20241018-053121-57723cc2/meta/               0                 0
## 3    20241018-053121-57723cc2/meta/config.rds            1460              1455
## 4      20241018-053121-57723cc2/meta/info.rds            3417              3412
## 5  20241018-053121-57723cc2/meta/manifest.rds             286               281
## 6   20241018-053121-57723cc2/meta/session.rds           24882             24877
## 7              20241018-053121-57723cc2/pack/               0                 0
## 8  20241018-053121-57723cc2/pack/incoming.csv             542               888
## 9   20241018-053121-57723cc2/pack/orderly.yml             214               358
## 10     20241018-053121-57723cc2/pack/script.R             175               220
##              timestamp permissions    crc32 offset
## 1  2024-10-18 05:31:20         755 00000000      0
## 2  2024-10-18 05:31:20         755 00000000     55
## 3  2024-10-18 05:31:20         644 75a64269    115
## 4  2024-10-18 05:31:20         644 afa9c90b   1661
## 5  2024-10-18 05:31:20         644 12a70cc5   5162
## 6  2024-10-18 05:31:20         644 9dbf364b   5536
## 7  2024-10-18 05:31:20         755 00000000  30505
## 8  2024-10-18 05:31:20         644 e3c70810  30565
## 9  2024-10-18 05:31:20         644 93fda3f3  31195
## 10 2024-10-18 05:31:20         644 58329b6b  31496

The subdirectory pack contains the report working directory, all code and dependencies, etc, while meta contains additional information required to run the report. In particular the incoming.csv file in the pack directory contains the dependency imported from other.

Then copy this zip file somewhere else to run it (details vary based on your system, and moving the file is not necessary to run it, though it will be the most likely situation).

Once the files have been moved we can run it with:

workdir <- tempfile()
res <- orderly::orderly_bundle_run(bundle$path, workdir)
## [ start      ]  2024-10-18 05:31:21.520194
## 
## > d <- read.csv("incoming.csv", stringsAsFactors = FALSE)
## 
## > png("graph.png")
## 
## > par(mar = c(15, 4, 0.5, 0.5))
## 
## > barplot(setNames(d$number, d$name), las = 2)
## 
## > dev.off()
## png 
##   2 
## 
## > info <- orderly::orderly_run_info()
## 
## > saveRDS(info, "info.rds")
## [ end        ]  2024-10-18 05:31:21.536892
## [ elapsed    ]  Ran report in 0.01669765 secs
## [ artefact   ]  graph.png: 67b3e662440b3978ee78b9d4cc159884
## [ ...        ]  info.rds: 0b29b8285cf56e1cc7084b4e07609a19

With the workdir being a directory that you want the report to be run in. This can be the same as the path the incoming zip file is found, if you want, but this will make it harder to know what has been run already or not.

This creates another zip file, but this time contains the results of running the report.

The result can be imported into order by using orderly::orderly_bundle_import with the path to the zip file:

orderly::orderly_bundle_import(res$path, root = path)
## [ import     ]  use_dependency:20241018-053121-57723cc2

The copy of use_dependency is now in the archive and can be used like any other orderly report

orderly::orderly_list_archive(path)
##             name                       id
## 1          other 20241018-053120-f01c0175
## 2 use_dependency 20241018-053121-57723cc2
orderly::orderly_graph("other", root = path)
## other [20241018-053120-f01c0175]
## └──use_dependency [20241018-053121-57723cc2]

Limitations

  • We do not yet verify that the incoming bundle comes from a trusted source, nor that the completed bundle was run on a trusted system, nor that the bundle pack was not modified en route. We will add a signing and verifying step in future that will address these issues.

  • We do not support encryption of the bundle, but may do so in a future version.

  • Secrets will be copied in clear text if included.

  • We do not check that the package versions on the remote machine are suitable, though metadata is included to support this in future.

  • We do not support reading from databases using the connection: field while the report is running (reading data before is fine). In future we may relax this so that the remote report interacts with the database. Practically this means that bundled reports cannot use the connection: field, and all data will be packaged into the bundle, and so may be large.

  • As discussed above in the overview section, orderly does not deal with the movement of bundles between machines, nor queuing of these bundles.