tagit is a command-line tagging system for experiments, focusing on getting rid of
researchers' waste of time on arranging / manipulating / moving the experimental result.
tagit is distributed via PyPI with the name of etagit.
The name of the script is the same as tagit, though.
$ pip3 install etagitsqlite3.h. Otherwise, tagit may output some errors. Instruction for some distros is given below.
$ sudo apt install libsqlite3-dev$ sudo yum install libsqlite3x-develConsider a curious sysadmin intern, Lisa, trying to discover the effect of system configurations on the performance (i.e. throughput) of server workloads.
Lisa measures the performance by changing system configurations: OS, memory size, storage device, network, database and the number of worker threads.
Since the total number of configuration sets is large (6 configurations with 3 different values each result in 729 combinations!),
Lisa writes a simple script to record the experiment result hierarchically;
configuration values are saved as directory trees (e.g., ubuntu1804/8/HDD/10/mysql/16/result.txt).
After all combinations are tested, Lisa wants to analyze the results using spreadsheet, however,
only the raw data is saved under directory structure.
Lisa googles out and notices that she should do the two jobs:
convert each data to useful data, and accumulate them into one file (e.g., .csv) that spreadsheet apps can import.
Lisa spends a day to learn utilities needed for extracting like sed, awk and grep,
and she finally makes it to import the data to spreadsheet.
After a month, Lisa decides to present her experiment as it is the last week of the internship. Unfortunately, she has lost the spreadsheet and what she got is only the raw data itself, but now Lisa has no problem in using various utilities. However, Lisa now finds a serious problem: what does '8' mean in the directory path?
# 1. Record the data
$ ./run_exp.sh | tagit record perf "storage=sata_ssd, mem=16GB"
IOPS is 20K
latency is 100us
New experiment: [perf]
[perf] New tag added:
- storage
- mem
$ ./run_exp.sh | tagit record perf "storage=nvme_ssd, mem=16GB"
IOPS is 40K
latency is 10us
$ ./run_exp.sh | tagit record perf "storage=nvme_ssd, mem=32GB"
IOPS is 60K
latency is 10us
# 2. List experiments
$ tagit list
perf
# 3. Report the data
$ tagit report perf
[perf] (storage=sata_ssd, mem=16GB)
- raw: IOPS is 20K
latency is 100us
[perf] (storage=nvme_ssd, mem=16GB)
- raw: IOPS is 40K
latency is 10us
[perf] (storage=nvme_ssd, mem=32GB)
- raw: IOPS is 60K
latency is 10us
$ tagit report perf "mem=16GB, storage"
[perf] (mem=16GB, storage=sata_ssd)
- raw: IOPS is 20K
latency is 100us
[perf] (mem=16GB, storage=nvme_ssd)
- raw: IOPS is 40K
latency is 10us
# 4. Parse recorded data
$ tagit parse add perf iops "awk '/^IOPS/{print \$NF}'"
[perf] New data category:
- iops
$ tagit parse add perf latency "awk '/^latency/{print \$NF}'"
[perf] New data category:
- latency
$ tagit parse list perf
id rule src dest updated
---- --------------------------- ----- ------- ---------
0 awk '/^IOPS/{print $NF}' raw iops True
1 awk '/^latency/{print $NF}' raw latency True
$ tagit report perf -d "latency, iops"
[perf] (storage=sata_ssd, mem=16GB)
- latency: 100us
- iops: 20K
[perf] (storage=nvme_ssd, mem=16GB)
- latency: 10us
- iops: 40K
[perf] (storage=nvme_ssd, mem=32GB)
- latency: 10us
- iops: 60K
# 5. Delete a data
$ tagit data remove perf "mem=32GB"
$ tagit report perf -d "latency, iops"
[perf] (storage=sata_ssd, mem=16GB)
- latency: 100us
- iops: 20K
[perf] (storage=nvme_ssd, mem=16GB)
- latency: 10us
- iops: 40K
# 6. Delete an experiment
$ tagit exp remove perf
$ tagit list
record$ tagit record <exp_name> <tags> [-d <data_category>] [-q]Records stdin tagged with <tags> in <exp_name> space.
<tags> must be specified with double quotes (e.g., $ ./run_exp.sh | tagit record myexp "a=1, b=2").
The data category into which data is recorded can be specified using -d option; the default category is raw.
If -q is specified, data is recorded quietly.
report$ tagit report <exp_name> [<tags>] [-c <csv_file>] [-f <path>] [-d <data_category>]Reports the data in <exp_name> space.
If <tags> are specified, tagit reports data corresponding to the specified tags.
By default, the result is printed in terminal.
Data category to be reported can be specified using -d option, and all data categories are reported by default.
Multi-valued tags
|."color=red|blue, shape=cube"Tag order
<tag>. For example, "c=3, b=2, a=1" makes tags printed in c-b-a order."c, b=2, a".
This is for setting the tag order without changing the report scope.Result format options
-c: Print the result in csv format-f: Print the result in file hierarchy; each data is saved as a file under the nested directory path, where each directory corresponds to a tagparseparse add$ tagit parse add <exp_name> [-s <src_category>] <dest_category> <rule>Add a parsing rule to an experiment <exp_name>.
For each data in experiment <exp_name>, <rule> is applied to raw category, and the result is saved in <dest_category>.
If <src_category> is specified, the <rule> is applied to <src_category>.
The <rule> should follow shell command format (For example: awk '^latency/{print \$NF}').
parse list$ tagit parse list <exp_name>List parsing rules in the experiment <exp_name>.
parse remove$ tagit parse remove <exp_name> [-a] [<rule_id>]Remove parsing rule of id <rule_id>: from the experiment <exp_name>.
If -a is specified, all parsing rules are removed.
list$ tagit list [<exp_name>]List the experiments. If <exp_name> is specified, it lists the name of tags and data categories in the specified experiment instead.
expexp rename$ tagit exp rename <exp_name> <new_name>Rename the experiment <exp_name> to <new_name>.
exp remove$ tagit exp remove <exp_name>Remove the experiment <exp_name> and its parser.
exp clean$ tagit exp clean <exp_name>Clean all data from the experiment <exp_name>.
datadata remove$ tagit data remove <exp_name> <tags>Remove data from the experiment <exp_name> by tags.
tagtag rename$ tagit tag rename <exp_name> <name> <new_name>In the experiment <exp_name>, rename the tag <name> to <new_name>.
tag fix$ tagit tag fix <exp_name> <tags>In the experiment <exp_name>, fix the value of tags according to <tags>.
For example, "color=blue, shape=cube" fixes the tags color and shape to blue and cube, respectively.
When the value of the tag fixed by this command is not specified, the value is set to the fixed value.
The fixed values can be overrided by values specified during recording.
tag unfix$ tagit tag unfix <exp_name> <tags>In the experiment <exp_name>, unfix the value of tags according to <tags>.
Tags must be specified without values in <tags> (e.g., "color, shape").
tag update$ tagit tag update <exp_name> <tags>In the experiment <exp_name>, update the value of tags according to <tags>.
In <tags>, = is used for limiting the scope of update, and -> is used for specifying the value to update.
For example, "color=red, shape=sphere, color->blue, volume->10L" updates color to blue and volume to 10L, for records with color=red, shape=sphere.
export$ tagit export <output_dump>Export all experiment and data to <output_dump>.
import$ tagit import <db_dump>Import experiments and data from <db_dump>.
reset$ tagit reset [-y]Reset tagit database. -y passes all yes to prompts automatically.
tagit has five main concepts: experiments, tags, data, data categories, and parsing rules.
tagit).tagit.
Each data belongs to a data category, and may have multiple tags (e.g., "color=red, shape=sphere").raw by default,
and the data can be parsed and saved into different categories (e.g., throughput, latency).
Experimental results or stats are generally recorded in a single place (a simple example is /proc/vmstat, which shows the stats of virtual memory subsystem),
thus data categories can be useful for parsing necessary information from the whole data.This tool is currently a beta version and may contain some bugs. Issues & pull requests about new features, bug reports are appreciated.