# Short links, one per customer

M1 · Three copies, one campaign

You know what three copies of your step 2 service do with a campaign's 400,000 links.

• ○ D1 R1 , not recorded yet

  Before changing anything, run three copies of your step 2 service behind one address, as in production, and play the campaign tool: links for https://pantry.test/offers?c=<n>&coupon=xmas-<n>, n from 1 to 400,000. Keep each code with its address. Note the time, failures and shared codes.

  expectYou have the baseline: time, failures and shared codes for the service you shipped in step 2. Nothing sets a target for it. If it stalled, write down how far it got and what it was waiting for.

M2 · A code for every customer

A campaign's links are made in half an hour, no code belongs to two customers, none can be guessed from another, and a retry makes nothing new.

1. R1
   With the service running as three copies behind one address on your machine, set up as it runs in production, the campaign tool creates 400,000 links for 400,000 different addresses in the half hour before the SMS goes out, and no request fails after the tool's retries.
2. R2
   No code ever belongs to two links, however many copies create links at the same time. Every code is at most eight letters and digits and opens exactly the address it was made for.
3. R3
   Changing one character of a code a customer holds does not open anyone else's link.
4. R4
   Asking for an address that already has a link gives back that link, even when the request reaches every copy at the same moment.

• ○ D2 R1 , not recorded yet

  After your change, run D1 again with 400,000 addresses no link has yet (a new coupon prefix and campaign), with the same 100 workers and retries, against the same three copies.

  expectNo request failed after its retries. Say how long it took next to the tool's half hour, and what limited the rate on your machine.

• ○ D3 R2 , not recorded yet

  From D2's record, count the codes that came back for more than one address, and check that every code is at most eight letters and digits. Then open 2,000 of them at random through the address in front of the copies, without following redirects, and compare each Location with its address.

  expectNo code came back for two addresses, every code has the right shape, and all 2,000 open exactly their own address.

• ○ D4 R4 , not recorded yet

  Take 1,000 addresses no link has yet, in a campaign of their own. For each, send the same create request to all three copies directly at the same moment, 20 addresses at a time. Then read that campaign's numbers.

  expectAll three copies gave each address the same code, and the campaign has exactly 1,000 links.

• ○ D5 R3 , not recorded yet

  Pick 200 codes from D2's record at random. For each, request every code that differs from it in exactly one position, trying there each other character your codes can use, without following redirects. Count the ones that redirect.

  expectNone of them redirects. Work out first how many hits pure chance would give with your number of links and possible codes: if that is not far below one, your codes are too easy to guess.

M3 · Maya's number, live

The team reads how many customers opened a campaign, and the public service still does everything steps 1 and 2 promised.

1. R5
   The team can read, for one campaign, how many links it has, how many were opened at least once and how many openings there were, while Maya waits, with 400,000 links in the campaign.
2. R6
   What steps 1 and 2 promised still holds with three copies: links made before this step redirect, every redirect is counted, a disabled link answers 410 within ten seconds, only the team creates links, reads counts and disables, and the service runs at its public HTTPS address.

• ○ D6 R5 , not recorded yet

  A minute after D3, read the campaign's numbers. Through the address in front of the copies, open 10,000 of its links D3 did not open, and 1,000 of those again, 50 at a time, without following redirects. A minute later, read the numbers again and time the request.

  expectThe campaign still has 400,000 links, opened at least once went up by exactly 10,000 and openings by exactly 11,000. Say how long the read took and what it had to touch.

• ○ D7 R6 , not recorded yet

  Through the address in front of the three copies, open another of the campaign's links 60 times, then disable it. From ten seconds after the disable, open it 60 more times, and read its count a minute later.

  expectThe first 60 redirect, all 60 after the ten seconds answer 410, whichever copy took them, and the count is exactly 60.

• ○ D8 R6 , not recorded yet

  On your public address, play the campaign tool for 1,000 new addresses in a campaign of their own, and count codes shared by two addresses. On a phone on mobile data, open one of them and a link from step 1. Then, without the team's credential and with a wrong one, try to create a link.

  expect1,000 links with 1,000 different codes, both links land on their address over HTTPS, and both attempts without the credential are refused.