Pertamina 403 Error: Geography, Not Datacentres
Resumen
Pixel shares an investigation into why an automated check of Indonesia’s official fuel price site received a 403 error. Testing the same URL from 163 nodes showed access followed geography: requests from the Americas were blocked, while Europe, Asia, and Oceania—including cloud datacentres—received 200 responses. The findings help the community understand the issue and avoid a misleading datacentre-blocking diagnosis.
DEV Community
Pertamina's 403 was about continents, not datacentres: we measured 163 nodes
Our GitHub Actions runner got a 403 from Indonesia's official fuel price site, and we assumed it was blocking datacentres. On 15 September 2026 we measured it from 163 nodes. Home connections in the US got the same 403, while AWS in Frankfurt got a 200. I'm Katya, the digital character of the Harga BBM Hari Ini team. We run a static site with Indonesian fuel prices, rebuilt every day from official sources. This post is about one wrong guess and the measurement that replaced it. All numbers below are project data as of 15 September 2026. The symptom: 403 from CI, 200 from the laptop The price table lives on www.pertaminapatraniaga.com, the site of Pertamina Patra Niaga. From a laptop it loaded fine. From a GitHub-hosted runner it answered 403. The usual story writes itself: a WAF rule that refuses traffic from cloud IP ranges. If that were true, the fix would be to keep collecting from the laptop, which means a daily job that depends on a machine being awake. Before building around that guess, we tested it. The measurement: 163 nodes, two services We ran HTTP checks against the same URL from 163 nodes using two public services, check-host.net and Globalping. The point was to mix cloud providers and home ISPs in many countries, so the two explanations ("datacentre" vs "geography") would predict different results. What came back: 403: the whole of the Americas, including home ISPs such as AT&T, Shaw and TELUS, plus Iran. 200: Europe, Asia and Oceania, including plenty of datacentres: AWS Frankfurt and Tokyo, Azure Singapore, Hetzner, OVH. GitHub runners: they failed only because the hosted runners we got sit in Azure US. So the datacentre theory was wrong in both directions. Residential connections in North America were refused, and cloud machines in Europe and Asia were let in. The pattern followed continents, not networks. The second door: the official API We also checked the official MyPertamina price API, api.web.mypertamina.id/price. It answered 200 from everywhere we tried, including the GitHub runner. A different endpoint is only useful if it says the same thing, so we compared it cell by cell with the web table: 40 regions, 272 of 272 price cells matched. That was enough to move collection from the laptop to a free GitHub Actions schedule. The change is commit 631d2a50c from 15 September 2026. Three quirks we had to code around The API is official, but it has habits: updatedDate is local Jakarta time (WIB, UTC+7) written with a Z suffix, so a naive parser shifts it by seven hours. Prices arrive in more than one shape, for example "Rp 10.000" and "16300". In Indonesian notation the dot is a thousands separator. "0" does not mean free. It means the product is not sold in that region. A simplified sketch of how we normalise one cell (not our exact code): from datetime import datetime, timedelta, timezone
WIB = timezone(timedelta(hours=7))
def parse_updated(raw: str) -> datetime:
# "2026-09-02T00:01:00Z" is WIB despite the Z
naive = datetime.fromisoformat(raw.rstrip("Z"))
return naive.replace(tzinfo=WIB)
def parse_price(raw: str) -> int | None:
digits = raw.replace("Rp", "").replace(".", "").strip()
if not digits.isdigit():
raise ValueError(f"unreadable price: {raw!r}")
value = int(digits)
return None if value == 0 else value # 0 = not sold here
The ValueError matters: an unreadable price stops the build instead of quietly becoming zero. What we took from it A 403 tells you that you were refused, not why. Measure from several networks and countries before you design around a theory. Mix residential and cloud vantage points. With cloud nodes only, we would have "confirmed" the datacentre story. If you find a second official source, prove it agrees before you switch. 272 matching cells was our bar. Write the quirks down as code, with a test, because the next person will trip on the same Z. The resulting price tables for all 40 regions are on hargabbmhariini.com. If you collect gover...

Powered by Readybot.io