Commit graph

3 commits

Author SHA1 Message Date
Frederic Jung
dd6e4f4f8a
Fix CodeQL code-scanning alerts
- Values pasted into eval'd or inline-script code go through lit() (scripts/lib/js-literal.js), which escapes <, >, /, U+2028 and U+2029 after JSON.stringify
- build test and i18n-check read script tags with a DOM parser or a plain search, not HTML regexes
- mcp.js parses the Authorization header without a backtracking regex
2026-11-05 21:24:26 +01:00
Frederic Jung
ad9aabb265
Follow-ups to #271
- IX ports are measured at their nominal capacity: a LACP fault or one
  building down no longer buys a bigger port for a day or two.
- The order stops before working anything out when its budget is spent
  (31 ms a day to 0.05 ms on the big network at 3x), and a port that takes
  a LAG no longer prices every upgrade first.
- The log names the exchange, not its city.
- Tests for the LACP case and a remote IX port; a test comment's numbers;
  a shorter CHANGELOG bullet that no longer promises a per-POP switch for
  remote ports.
2026-11-04 08:03:10 +00:00
Christopher Hawker
62c242c9a7 The intern can keep PNI and IX ports ahead of their traffic (closes #234)
A new standing order, Keep peering headroom, works like the transit one: it
keeps each PNI and IX port above a multiple of today's traffic (1.25x to 3x).
A single port at a POP with several routers gets a second port in a LAG; any
other port moves up to the smallest port size that fits, or the biggest the
router takes. If the preferred move cannot be done it tries the other one.

It pays through the same actions as a click (ACT.pni, ACT.lagP, ACT.buyIx,
ACT.lagIx), is off by default, has the Headroom and Budget dials and the
per-POP switch, and keeps to the cash floor. A PNI augment on order is waited
for, router ports an augment will take are not promised twice, and nothing
is bought while a DDoS flood is live, since the flood counts as port usage.

The intern's monthly bill figure (opsMoNow) now counts IX ports too.
2026-11-03 22:54:54 +11:00