Documentation for 3.14
Other versions: trunk  4.3  4.0  3.27  Older...

Running TextTest Unattended
A Guide to Batch Mode

Introduction
It can be very useful to have TextTest run lots of longer tests (say) overnight and provide the results in an email or HTML report rather than have one of the interactive user interfaces present. That is the purpose of “batch mode”. To select batch mode, provide the command line option “-b <batch_session>” or fill in the “Run Batch Mode Session” tab under “How to Run” in the static GUI. In general, you will probably want to start such batch runs via a script, for example using crontab on UNIX.
The batch mode “session” is simply an identifier that defines a particular sort of batch run. Most of the batch mode configuration can be defined per session. Any identifier at all can be provided, and if no configuration is recognised for that session name default settings will be used. All of the batch mode config file settings that start with “batch_” (described below) are “composite dictionary entries” with the batch session names as keys, it is recommended to read the file format documentation for what this means.
The email report and where it is sent
TextTest batch mode generates an email report. For a multiple-developer project it is often useful to direct such reports to a newsgroup, providing everyone the chance to see at a glance what works and what doesn't. This will then generally look something like this (example newsgroup viewed in mozilla):



As you see from this example, the title of the mail consists of the date and a summary of what tests were run and what happened to them (for the application “Tail”, in this case). If the -name option is provided to the run on the command line, that name is used to define the run instead of the date. (In general, use -name to test actual named releases, and the default date-functionality with nightjobs)
The body of the mail contains two further sections, one which summarises exactly which tests failed and a further section which endeavours to give some details as to why they failed. These sections can be explored or ignored depending on how involved the reader is in the project. Managers will generally only need to look at the subject lines...
The name of the application, as provided here, can be configured via the config file entry “full_name”. By default a capitalised version of the file extension used for the application will be used here, but this doesn't always look so nice in reports.
The “details” section consists of the textual previews as generated by the dynamic GUI's “text info” tab when a test fails. It can be configured in the same way. In addition, it can be useful to configure the maximum width of lines allowed: some newsgroups have maximum line length limits and you don't want test reports bouncing. This can be done via the config file entry “max_width_text_difference”.
You can also configured it so that mail is only sent if at least one test fails, set the config file entry “batch_mail_on_failure_only” to "true". By default mail will always be sent irrespective of what happens.
Where the mail is sent to is controlled by the config file entry “batch_recipients”. This can be configured per batch session, and may be a comma-separated list for multiple recipients. The sender address can be controlled by the “batch_sender” config file entry, while the SMTP server to use for sending mail can also be configured via “smtp_server”.
All of these will need to be configured on Windows as no defaults are provided. On UNIX, the SMTP server defaults to “localhost” and both sender and recipient addresses default to “$USER@localhost”, so it is generally only necessary to configure the recipients.
Generating HTML reports
For more flexibility in viewing and analysing a lot of results, as well as being able to easily monitor the behaviour of particular tests over time, it can be very useful to store the batch results in a repository and generate HTML reports from them. In order to store the information from the batch runs, the config file entry “batch_result_repository” should be set to a directory under which batch results can be stored. Results are then stored per test and day and are never overwritten: to recreate results for a particular day it is necessary to explicitly remove the previous ones, either manually or via the archiving script described below.
For the location of the actual reports, set the config file entry “historical_report_location” to another directory. Both of these are composite dictionaries as described above so both can be varied per batch session. In order to actually generate the report, run with the -coll flag (see below) which will rebuild all the reports from scratch based on what is in the repository.
By default, each application will be shown on a separate page named after that application. Sometimes though it can be useful to have several applications on the same page, especially if they don't have so many tests each. In this case you can set the config file entry "historical_report_page_name": the name given will be used as the title of the HTML report and all applications with the same value will appear on the same page.
The easiest way to get a handle on what this looks like is to look at this small example, which is generated by TextTest's tests for itself. Each day's results correspond to a column, while each test has a row. The results can be explored by clicking around. For a real such site in active use, check out the nightly runs of TextTest's self tests page, which is updated every day with the latest status.
The colours in the site are also configurable: use the config file dictionary setting “historical_report_colours”. To see how to set this, look at the config file table and pattern match on the default value. Note that you can now configure the run header colours per day of the week if you wish to highlight runs made, for example, at the weekend when more tests may be run.
Configuring which subpages are produced in the HTML report
Up to TextTest 3.12 there were two "subpages" produced as part of the HTML report: one showing the last six runs and one showing all of the runs. This is now completely configurable. The names of the subpages produced are controlled by the config file setting "historical_report_subpages", which by default will contain only "Last Six Runs" which works as before. It is however easy to remove this page via the {CLEAR LIST} mechanism.
Such a subpage will contain a subselection of the runs present in the repository. If no subselections are specified, all the runs will be included. So restoring 3.12's "All" page is just a matter of adding:
historical_report_subpages:All
There are two ways of subselecting the tests. You can specify a cutoff point, similar to "Last Six Runs", by specifying the dictionary entry "historical_report_subpage_cutoff". This should be keyed on the name of the subpage specified as above, and the value is the number of runs to include. You can also filter on days of the week, by providing a list of weekdays to include in the entry "historical_report_subpage_weekdays", which is also keyed on the name of the subpage. So for example, if we wanted to include a page with the last fortnight's results and a page only showing runs done at the weekends, we could do:
historical_report_subpages:Last Fortnight
historical_report_subpages:Weekend

[historical_report_subpage_cutoff]
Last Fortnight:14

[historical_report_subpage_weekdays]
Weekend:saturday
Weekend:sunday
Archiving old data from the HTML reports
After a while, very old test results in the repository cease to be interesting and can safely be archived. This is done via the script batch.ArchiveRepository, with arguments 'after' and 'before' for the time period to archive (and 'session' for the batch session to do it on, defaults to all known sessions). The dates should be in the same format as the dates on the pages, e.g. 21Jan2005. This should be less critical now if you aren't generating an "All" page, which otherwise can take an extremely long time to generate with a year's data in it, but even so just searching for the files can take a while if there are a lot of them. It doesn't actually delete things but moves them to an <app>_history subdirectory of the location specified by "batch_result_repository", where they will not be picked up by report generation script and may be restored if desired.
Configuring what tests are run by batch mode sessions
The config file entry “batch_timelimit”, if present, will run only tests which are expected to take up to that amount of time (in minutes). This is of course only useful if performance testing is enabled for CPU time.
More generically, you can use the “batch_filter_file” entry to identify filter files to be associated with a particular batch run. These can either contain a list of tests or search criteria to apply and can be edited using the static GUI. In this context it can be useful to note that such filter files can contain application and version-specific suffices in case similar criteria imply different tests for different applications: this allows the same entry for batch_filter_file to indicate different tests for different applications and versions.
If certain versions should be run automatically as part of a batch mode run without needing to explicitly specify them on the command line, the entry “batch_extra_version” can be used for this purpose. This is a more specialised version of the “extra_version” setting. For the purposes of configuring which tests should be included in such versions it is also possible to start the GUI with these versions present, use the command line option "-bx <batch session name>" to achieve this.
If the entry “batch_use_version_filtering” is set to “true”, all versions are assumed to be disabled unless explicitly enabled by being included in the “batch_version” list setting. The point of this is in the presence of multiple test applications and multiple releases of the system: a single run of TextTest can be started with a particular version identifier and each application can decide in its config file if it wants to run tests for that version of the system. This is generally easier than trying to set up separate nightjob runs for each application.
Both of these things act in concert with any test selection filters selected on the command line or from the static GUI. As described there, only tests which satisfy all filters present will be selected.
Collecting multiple emails into a single one
When many versions of the system under test are active, and many different hardware platforms are used, you may want to test the system on all of these combinations. This can lead to a great many test runs and consequently a lot of emails. It is often easier to read these if they are collected into a single larger email: otherwise it is hard to get an overview of what is happening.
To do this, set the config entry “batch_use_collection” to "true" for the batch session in question. This will ignore the email-sending settings and send the batch report to an intermediate file. When all tests have been run in this way, run the collection script by providing the "-coll" argument on the command line. This will also build the HTML report described above, if that is enabled. To build only the email report or only the HTML report, you can also run with "-coll mail" or "-coll web" respectively. (The older plugin script "batch.CollectFiles" will still work but is now deprecated)
TextTest will then search for all such intermediate files and amalgamate them into a single mail per application. If a version is provided to this script via -v <version>, only runs which ran with that version identifier will be collected. This applies to the HTML report as well.
By default this collection procedure will find all batch runs that have currently stored their results under the root temporary directory. There are two ways in which this can be configured to handle errors in the processes that are supposed to create these runs.
For a start, you can set the config entry "batch_collect_max_age_days" which will ensure that no batch run older than the given number of days will be picked up and reported. This is useful in case the expected batch run has not been run for some reason, otherwise the results of the previous run may be picked up and lead to confusion.
Also, you can tell it what it should expect to find. By providing the entry "batch_collect_compulsory_version" (which is a list), you can tell it to report an error if a batch run for a certain version was not found, thus making it clearer if something has not run when it was supposed to. Both these settings are keyed on batch session names as described above.
Example config file


This config file configures TextTest so that:
  • “-b local” will send email to the sender directly
  • “-b nightjob” will run all tests that take up to 180 minutes (3 hours). It will only accept the version identifiers “11”, “12” and “linux” (if we also set “batch_use_version_filtering” to true, which we didn't here...) It will write its results to an intermediate collection file, where they can be collected later.
  • “-b wkendjob” will run all the tests that there are. It will also accept the version identifiers “sparc” and “powerpc” (evidently we want to test on more flavours of UNIX at the weekends!). It will use collection in a similar way to the “nightjob” session.
  • All of these will write their results under the location /some/central/directory, provided no previous results have been calculated that day. When the collection is run, the files from the 'nightjob' or 'wkendjob' sessions will be amalgamated and mailed to the carmen.test_newsgroup mail address. The website at /our/documents/html/testreports will also be regenerated from scratch from the repository described above.
Reconnecting the User Interfaces to batch results
Batch mode's email report is all very well, but alone it doesn't give you the power of the GUI to view results in detail or to save them if that would be appropriate. It can be very useful to “reconnect” the GUI as if a batch run had been run using it. To do this, select “Reconnect” from the Actions menu in the Static GUI, which will bring up a dialog where you can browse your file system to find the files for the run (or provide the “-reconnect <directory>” option on the command line to the dynamic GUI or console interface).
There are essentially two ways to select previous runs. Either you explicitly select the run directory (it will be called something like "hello.version.13Aug171624.9856"), or you leave the parent directory selected and provide a version filter in the “Version to Reconnect to” field. This last possibility will then load in all runs that were run with that version identifier, or some more specific version identifier: i.e. if you provide "foo" it will load runs for "foo.bar" and "foo.blah" also if they exist. Not selecting either will load in every previous run that has left files in the relevant temporary space.
There is a switch at the bottom which allows you to choose between a quick re-display of what was displayed in the original batch email report, and an option to recalculate the results from the raw files. If for any reason the quick re-display isn't possible, it may trigger a recomputation anyway.
The recomputation, whether explicitly requested (-reconnfull on the command line has the same effect) or auto-triggered, will take the raw output of the run reconnected to and reapply the text filtering mechanisms to it, and also re-evaluate any automatic failure interpretation that could be triggered. This is useful if you have updated your config file filters in the meantime and want to see if they are applied correctly. It is, of course, a good deal slower than simply re-reporting what was present before, as occurs by default.


Last updated: 05 October 2012