Multi-valued by design
Uniqueness is on the triple, not the pair — so a node can carry two VLANs, three networks, any repeated key, with nothing clobbered.
HPC · cluster node attributes
The genders flat file, made relational.
Classic genders is a flat (node, key)=value bag — no
multi-values, no real relationships, no query language. gendersdb
puts SQLite where genders kept a text file: an (entity, name, value)
triple model plus typed relations. Your nodeattr and pdsh
muscle memory still works — everything else gets an upgrade.
the model, fixed
It fixes the model, instead of wrapping it.
The flat file couldn't say a node has two VLANs, or that a rack is powered by a phase-B PDU. gendersdb makes attributes multi-valued and relationships first-class — so the questions you always wanted to ask become one line.
/etc/genders · flat
# one (node,key)=value bag; last write wins node[01-08] compute,eth0=e%n,bmc=%n-ipmi node[01-04] gpu node01 vlan=10 node01 vlan=20 # clobbers 10 # racks? PDUs? phases? nowhere to put them.
- single value per key
- no relations between nodes
- substring / glob matching only
gendersdb · triple + relations
set node01 vlan=10 set node01 vlan=20 # both kept link node01 member_of rack-b link rack-b powered_by pdu-b set pdu-b phase=B
- native multi-valued attributes
- real 1‑n / m‑n edges (node→rack→PDU)
- booleans, numbers, regex, hops, aggregation
what you get
Small tool. Sharp edges in the right places.
Relations & inheritance
Typed edges model real topology — node→rack→PDU. Attributes inherit one hop along member_of, opt-in when you want it.
A real query language
Booleans, numeric comparison, regex, relational hops and aggregation — compiled to safe, parameterized SQL. compute && mem>=256 && !gpu.
Natural language
Ask in English; gendersdb translates to the DSL, echoes it, and waits for your yes. It proposes a verifiable expression — it never selects nodes on its own. Local, never cloud.
Interactive shell
A REPL over your database — tab-completion of live names, history, hostlist expansion, and staged begin/commit/rollback transactions.
Drop-in compatible
A nodeattr shim speaks the classic CLI, and export emits a flat genders file — so existing nodeattr/pdsh scripts run unchanged.
Head-node friendly
Standard library only — no pip install, no daemon, no cache. It drops onto a locked-down HPC head node with just Python 3.8+ and SQLite.
the DSL in anger
Ask the cluster hard questions.
Every value is a bound parameter; every attribute and relation name is charset-validated. The same grammar drives the CLI, the shell and natural language.
$ gendersdb select 'compute && mem>=256 && !gpu'
node05 node06 node07 node08 # booleans + numeric comparison
$ gendersdb select 'compute && @powered_by(phase=B)'
node0001 node0003 node0005 … # one relational hop — 1696 nodes on phase B
$ gendersdb agg sum cpu compute
57048 # capacity rollup across thousands of nodes
$ gendersdb nl "compute nodes on PDU phase B"
DSL: @powered_by(phase=B) && compute # echoed for you to confirm, then run
download · install · run
Up and running in three commands.
Download the package, unpack, and run the installer — no pip, no clone.
Already have an /etc/genders? Import it and keep going; a sample
flat file ships in the box.
Quick start
the migration story, end to end
-
1 · create the database
$ gendersdb --db cluster.gdb init -
2 · import your flat genders file
$ gendersdb --db cluster.gdb import /etc/genders -
3 · ask a question
$ gendersdb --db cluster.gdb select 'compute && !gpu' $ gendersdb --db cluster.gdb nl "gpu nodes with 512+ of ssd" -
or drop into the shell
$ gshell --db cluster.gdb
Download & install
v1.0 · per-user or system-wide; no pip
# unpack, then:
$ tar xzf gendersdb-1.0.tar.gz
$ cd gendersdb-1.0
$ ./install.sh # ~/.local
$ sudo ./install.sh # /usr/local
- ●Python 3.8 or newer
- ●SQLite (bundled with Python)
- ●Zero third-party packages
- ●MIT licensed · optional local Ollama for NL