Infinera DTN-X TL1 driver

The infinera_tl1 driver presents an Infinera DTN-X optical platform managed over TL1. It shares the TL1 core with Ciena and Cisco ONS (block reading, ACT-USER parsing, response framing) and differs everywhere the hardware does.

Two of those differences are why this driver exists. Both break clients written against Ciena, and neither is reproducible without a simulator that models them.

The prompt is >

Not <. A client that waits for < on a DTN-X waits until its read timeout.

RTRV-TIDMAP is paged

A DTN-X returns one logical answer as several blocks. Every block but the last is coded RTRV rather than COMPLD, and the prompt is written between blocks:

> ACT-USER::admin:1::"admin";
> RTRV-TIDMAP:::2;

   INF-SJC-1000 26-09-12 19:53:38
M  2 RTRV
   "::TID=STLTND1Y,NODEID=MA0353062311,ROUTERID=11.253.152.44"
   ... nine more records in this block ...
;

>
   INF-SJC-1000 26-09-12 19:53:38
M  2 RTRV
   ... ten more ...
;

>
   INF-SJC-1000 26-09-12 19:53:38
M  2 COMPLD
   "::TID=TUSTNN2Y,NODEID=MA8953593801,ROUTERID=11.253.152.64"
;

>

Ten records per block. The failure this reproduces is not a truncated list:

Why the prompts matter

A client that reads “until the prompt” gets block one, returns it as the whole answer, and leaves blocks 2..N sitting in the socket. Those then arrive as the reply to its next command, which fails CTAG correlation, and so does every command after it. One paged response silently desyncs the rest of the session.

Generated nodes carry 12 to 30 remote nodes, deliberately more than one page, so the paging path is always exercised. A fleet that never pages cannot regression-test the fix.

Routed replies carry the system name

A DTN-X relaying a command for another node answers under its own system name:

> RTRV-EQPT:STLTND1Y::3;

   INF-SJC-1000 26-09-12 19:54:26
M  3 COMPLD
   "CHASSIS-1::PROVISIONED,SN=FOCCBDD420D,SWVER=R21.3,IP=127.0.0.1:IS-NR"
   ...
;

The header says INF-SJC-1000 even though the command addressed STLTND1Y. A client cannot confirm routing by comparing the header SID to the TID it asked for, the way it can on a 6500. This is real behaviour, it looks like a bug, and TestInfineraRoutedResponseKeepsSystemName pins it so a well-meaning change cannot “fix” it into Ciena’s behaviour.

Records

"::TID=CSVLTNFCO1Y,NODEID=MA4623110007,ROUTERID=11.253.152.33"

Keyword fields with an empty AID. TIDs are opaque alphanumeric names, not readable site labels. ROUTERID is shaped like an address but is a routing identifier, so a client should not store it as a management IP.

Commands

Command	Behaviour
ACT-USER::user:ctag::pass;	In-band login; DENY PLNA before it succeeds
RTRV-TIDMAP	The remote nodes behind this one, paged
RTRV-EQPT	Chassis inventory, streamed zero-copy; routed by TID
RTRV-ALM-ALL	Active alarms
RTRV-COND-ALL	Standing conditions
RTRV-SW-VER	Software version
RTRV-SYS	System identity

Any other verb after login returns DENY with ICNV.

Generating Infinera devices

./bin/rcfg-sim-gen -count 2 -ip-count 1 -devices-per-ip 2 \
  -ip-base 127.0.0.1 -port-start 22400 \
  -distribution "infinera-dtnx-tl1:100" \
  -output-dir /tmp/inf -manifest /tmp/inf/manifest.csv

Manifest rows carry vendor=Infinera, template=infinera_tl1. Mix with other models in one run, e.g. --distribution "infinera-dtnx-tl1:50,cisco-ons15454-tl1:50".

See worked examples for full sessions.