| 1 | //! What g1t earns on each thing it sells, measured against what |
| 2 | //! Cloudflare actually charged for it. |
| 3 | //! |
| 4 | //! Once a day, after `costs` has read Cloudflare's bill, the reconciler |
| 5 | //! puts three figures side by side for every day and each of g1t's |
| 6 | //! products (a "bucket": sandboxes, deployments, git, repository storage, |
| 7 | //! …): |
| 8 | //! |
| 9 | //! 1. **What Cloudflare charged**: the day's cost lines `cost_map` gives |
| 10 | //! the bucket. |
| 11 | //! 2. **What g1t's meters recorded**: the cost on the ledger's entries for |
| 12 | //! it (the price book's cost at the time) and, where a mapping names |
| 13 | //! one, g1t's own count of the same units (git operations). |
| 14 | //! 3. **What customers were charged**: the entries' value at price, before |
| 15 | //! the plan's included usage, a trial or a pool paid part of it; and of |
| 16 | //! that, what workspaces paid. Month-end meters (git, storage, scans, |
| 17 | //! embeddings, the actions cache) come from daily snapshots of what they |
| 18 | //! had come to (`pending_days`). The plan's price is the `platform` |
| 19 | //! bucket's: the plan pays for running g1t. |
| 20 | //! |
| 21 | //! From those: margin per product (value against cost) and for all of g1t |
| 22 | //! (money in against every cost); drift (counts or costs that disagree past |
| 23 | //! a mapping's threshold, and leaks: cost with no revenue, or a Cloudflare |
| 24 | //! meter no one mapped); each workspace's cost, Cloudflare's figure shared |
| 25 | //! out by each workspace's own meters; and price proposals when a unit's |
| 26 | //! real cost has moved (`pricing`). Alerts go to staff by email and as a |
| 27 | //! banner in sudo. See docs/BILLING_OPERATIONS.md. |
| 28 | |
| 29 | use std::collections::{BTreeMap, BTreeSet}; |
| 30 | |
| 31 | use g1t_contracts::billing::*; |
| 32 | use g1t_contracts::{FailureCode, Outcome, new_id}; |
| 33 | use g1t_contracts::time::rfc3339; |
| 34 | use g1t_kit::now_ms; |
| 35 | use serde::{Deserialize, Serialize}; |
| 36 | use worker::wasm_bindgen::JsValue; |
| 37 | use worker::{Env, Result}; |
| 38 | |
| 39 | use crate::Billing; |
| 40 | use crate::costs::{self, ARTIFACTS_OPERATIONS, DAY_MS, Rule, SOURCE_ARTIFACTS, SOURCE_BILLABLE, UNMAPPED}; |
| 41 | |
| 42 | /// Buckets that are the cost of running g1t, paid by the plan rather than |
| 43 | /// sold by the unit: never a leak for having no revenue of their own. |
| 44 | pub(crate) const OVERHEAD: [&str; 1] = ["platform"]; |
| 45 | /// Buckets Cloudflare does not bill: their cost is g1t's own figure. |
| 46 | pub(crate) const NOT_CLOUDFLARE: [&str; 1] = ["models"]; |
| 47 | /// The days drift is judged over. |
| 48 | const DRIFT_DAYS: u64 = 7; |
| 49 | /// The days a workspace's cost is set against its revenue. |
| 50 | const ANOMALY_DAYS: u64 = 30; |
| 51 | /// The days a unit's cost is measured over. |
| 52 | const MEASURE_DAYS: u64 = 30; |
| 53 | /// Fewer of g1t's units than this say nothing about cost per unit. |
| 54 | const MIN_UNITS: f64 = 1_000.0; |
| 55 | /// An open alert is emailed again after this long. |
| 56 | const REMIND_MS: u64 = 7 * DAY_MS; |
| 57 | |
| 58 | // --------------------------------------------------------------------- |
| 59 | // The arithmetic, apart from the database so it can be tested. |
| 60 | // --------------------------------------------------------------------- |
| 61 | |
| 62 | /// One of g1t's products on one day. |
| 63 | #[derive(Clone, Debug, Default, PartialEq)] |
| 64 | pub(crate) struct ProductDay { |
| 65 | pub day: String, |
| 66 | pub bucket: String, |
| 67 | /// What Cloudflare charged g1t, in millionths of a dollar. |
| 68 | pub cf_cost_micros: i64, |
| 69 | /// What g1t's meters recorded it cost (the price book's cost). |
| 70 | pub own_cost_micros: i64, |
| 71 | /// What customers were charged for it at price, before what paid. |
| 72 | pub value_micros: i64, |
| 73 | /// Of that, what workspaces paid themselves. |
| 74 | pub cash_micros: i64, |
| 75 | /// Units Cloudflare counted and units g1t counted, where a mapping |
| 76 | /// says they are the same units. |
| 77 | pub cf_quantity: f64, |
| 78 | pub own_quantity: f64, |
| 79 | } |
| 80 | |
| 81 | impl ProductDay { |
| 82 | /// What it cost: Cloudflare's figure where Cloudflare bills it, else |
| 83 | /// g1t's own (models are billed by their providers, through the gateway). |
| 84 | pub fn cost(&self) -> i64 { |
| 85 | if NOT_CLOUDFLARE.contains(&self.bucket.as_str()) { self.own_cost_micros } else { self.cf_cost_micros } |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | /// A line of Cloudflare's bill, as stored. |
| 90 | #[derive(Clone, Debug, Deserialize)] |
| 91 | pub(crate) struct LineRow { |
| 92 | pub day: String, |
| 93 | pub source: String, |
| 94 | pub product: String, |
| 95 | pub meter: String, |
| 96 | pub quantity: f64, |
| 97 | pub cost_usd: f64, |
| 98 | } |
| 99 | |
| 100 | /// A count of g1t's own, as stored. |
| 101 | #[derive(Clone, Debug, Deserialize)] |
| 102 | pub(crate) struct OwnRow { |
| 103 | pub day: String, |
| 104 | pub meter: String, |
| 105 | pub workspace: String, |
| 106 | pub quantity: f64, |
| 107 | } |
| 108 | |
| 109 | /// What a workspace was charged for one key on one day. |
| 110 | #[derive(Clone, Debug, Default, PartialEq)] |
| 111 | pub(crate) struct UsageRow { |
| 112 | pub day: String, |
| 113 | pub workspace: String, |
| 114 | /// A ledger task (or `builds`), a month-end source, or `plan`. |
| 115 | pub key: String, |
| 116 | pub value: i64, |
| 117 | pub cash: i64, |
| 118 | pub cost: i64, |
| 119 | } |
| 120 | |
| 121 | /// One workspace's share of a product's cost on one day. |
| 122 | #[derive(Clone, Debug, PartialEq)] |
| 123 | pub(crate) struct WorkspaceDay { |
| 124 | pub day: String, |
| 125 | pub workspace: String, |
| 126 | pub bucket: String, |
| 127 | pub cost: i64, |
| 128 | /// What the workspace paid in cash. |
| 129 | pub revenue: i64, |
| 130 | /// What its usage was priced at, whoever paid for it. |
| 131 | pub value: i64, |
| 132 | } |
| 133 | |
| 134 | fn micros(dollars: f64) -> i64 { |
| 135 | (dollars * 1_000_000.0).round() as i64 |
| 136 | } |
| 137 | |
| 138 | /// Puts the day's bill, g1t's counts and what customers were charged side |
| 139 | /// by side, a row per day and bucket, and shares each bucket's cost out |
| 140 | /// to workspaces. |
| 141 | pub(crate) fn fold( |
| 142 | rules: &[Rule], |
| 143 | revenue_map: &BTreeMap<String, String>, |
| 144 | lines: &[LineRow], |
| 145 | own: &[OwnRow], |
| 146 | usage: &[UsageRow], |
| 147 | ) -> (Vec<ProductDay>, Vec<WorkspaceDay>) { |
| 148 | let mut days: BTreeMap<(String, String), ProductDay> = BTreeMap::new(); |
| 149 | let entry = |day: &str, bucket: &str| -> ProductDay { |
| 150 | ProductDay { day: day.to_owned(), bucket: bucket.to_owned(), ..ProductDay::default() } |
| 151 | }; |
| 152 | // Which of g1t's own meters count each bucket's units. |
| 153 | let mut own_meters: BTreeMap<&str, BTreeSet<&str>> = BTreeMap::new(); |
| 154 | for rule in rules { |
| 155 | if let Some(meter) = &rule.own_meter { |
| 156 | own_meters.entry(rule.bucket.as_str()).or_default().insert(meter.as_str()); |
| 157 | } |
| 158 | } |
| 159 | let mut events: BTreeMap<(String, String), f64> = BTreeMap::new(); |
| 160 | for line in lines { |
| 161 | let rule = costs::classify(rules, &line.product, &line.meter); |
| 162 | let bucket = rule.map_or(UNMAPPED, |r| r.bucket.as_str()); |
| 163 | let key = (line.day.clone(), bucket.to_owned()); |
| 164 | if line.source == SOURCE_ARTIFACTS { |
| 165 | // What Artifacts counted: operations only, and only where the |
| 166 | // bill does not count them itself. |
| 167 | if ARTIFACTS_OPERATIONS.contains(&line.meter.as_str()) { |
| 168 | *events.entry(key).or_default() += line.quantity; |
| 169 | } |
| 170 | continue; |
| 171 | } |
| 172 | let row = days.entry(key.clone()).or_insert_with(|| entry(&key.0, &key.1)); |
| 173 | row.cf_cost_micros += micros(line.cost_usd); |
| 174 | if line.source == SOURCE_BILLABLE && rule.is_some_and(|r| r.own_meter.is_some()) { |
| 175 | row.cf_quantity += line.quantity; |
| 176 | } |
| 177 | } |
| 178 | for (key, quantity) in events { |
| 179 | let row = days.entry(key.clone()).or_insert_with(|| entry(&key.0, &key.1)); |
| 180 | if row.cf_quantity == 0.0 { |
| 181 | row.cf_quantity = quantity; |
| 182 | } |
| 183 | } |
| 184 | // g1t's own counts of the same units, by bucket and by workspace. |
| 185 | let mut own_by: BTreeMap<(String, String), Vec<(String, f64)>> = BTreeMap::new(); |
| 186 | // Cloudflare's own count by workspace, where it gives one |
| 187 | // (`cloudflare_<bucket>`): the best way to share its cost. |
| 188 | let mut cf_by: BTreeMap<(String, String), Vec<(String, f64)>> = BTreeMap::new(); |
| 189 | for count in own { |
| 190 | if let Some(bucket) = count.meter.strip_prefix("cloudflare_") { |
| 191 | cf_by.entry((count.day.clone(), bucket.to_owned())).or_default().push((count.workspace.clone(), count.quantity)); |
| 192 | continue; |
| 193 | } |
| 194 | for (bucket, meters) in &own_meters { |
| 195 | if meters.contains(count.meter.as_str()) { |
| 196 | let key = (count.day.clone(), (*bucket).to_owned()); |
| 197 | days.entry(key.clone()).or_insert_with(|| entry(&key.0, &key.1)).own_quantity += count.quantity; |
| 198 | own_by.entry(key).or_default().push((count.workspace.clone(), count.quantity)); |
| 199 | } |
| 200 | } |
| 201 | } |
| 202 | // What customers were charged. |
| 203 | let bucket_of = |key: &str| revenue_map.get(key).cloned().unwrap_or_else(|| "models".to_owned()); |
| 204 | let mut value_by: BTreeMap<(String, String), Vec<(String, f64)>> = BTreeMap::new(); |
| 205 | let mut cost_by: BTreeMap<(String, String), Vec<(String, f64)>> = BTreeMap::new(); |
| 206 | let mut revenue: BTreeMap<(String, String, String), i64> = BTreeMap::new(); |
| 207 | let mut valued: BTreeMap<(String, String, String), i64> = BTreeMap::new(); |
| 208 | let mut active: BTreeMap<String, Vec<(String, f64)>> = BTreeMap::new(); |
| 209 | for u in usage { |
| 210 | let bucket = bucket_of(&u.key); |
| 211 | let key = (u.day.clone(), bucket.clone()); |
| 212 | let row = days.entry(key.clone()).or_insert_with(|| entry(&key.0, &key.1)); |
| 213 | row.own_cost_micros += u.cost; |
| 214 | row.value_micros += u.value; |
| 215 | row.cash_micros += u.cash; |
| 216 | value_by.entry(key.clone()).or_default().push((u.workspace.clone(), u.value as f64)); |
| 217 | cost_by.entry(key).or_default().push((u.workspace.clone(), u.cost as f64)); |
| 218 | *revenue.entry((u.day.clone(), u.workspace.clone(), bucket.clone())).or_default() += u.cash; |
| 219 | *valued.entry((u.day.clone(), u.workspace.clone(), bucket)).or_default() += u.value; |
| 220 | active.entry(u.day.clone()).or_default().push((u.workspace.clone(), u.value.max(u.cost) as f64)); |
| 221 | } |
| 222 | // Each bucket's cost shared out: by Cloudflare's own count per |
| 223 | // workspace, else by g1t's own count of its units, else |
| 224 | // by what each workspace was charged for it, else by what its usage |
| 225 | // cost; running g1t, and what no one mapped, by each workspace's share |
| 226 | // of all usage that day. |
| 227 | let mut shares: BTreeMap<(String, String, String), i64> = BTreeMap::new(); |
| 228 | for ((day, bucket), row) in &days { |
| 229 | let key = (day.clone(), bucket.clone()); |
| 230 | let weigh = |m: &BTreeMap<(String, String), Vec<(String, f64)>>| m.get(&key).filter(|w| w.iter().any(|(_, v)| *v > 0.0)).cloned(); |
| 231 | let weights = if OVERHEAD.contains(&bucket.as_str()) || bucket == UNMAPPED { |
| 232 | active.get(day).cloned() |
| 233 | } else { |
| 234 | weigh(&cf_by).or_else(|| weigh(&own_by)).or_else(|| weigh(&value_by)).or_else(|| weigh(&cost_by)).or_else(|| active.get(day).cloned()) |
| 235 | }; |
| 236 | for (workspace, micros) in attribute(row.cost(), &weights.unwrap_or_default()) { |
| 237 | *shares.entry((day.clone(), workspace, bucket.clone())).or_default() += micros; |
| 238 | } |
| 239 | } |
| 240 | let keys: BTreeSet<(String, String, String)> = shares.keys().chain(revenue.keys()).cloned().collect(); |
| 241 | let workspaces = keys |
| 242 | .into_iter() |
| 243 | .map(|(day, workspace, bucket)| WorkspaceDay { |
| 244 | cost: shares.get(&(day.clone(), workspace.clone(), bucket.clone())).copied().unwrap_or(0), |
| 245 | revenue: revenue.get(&(day.clone(), workspace.clone(), bucket.clone())).copied().unwrap_or(0), |
| 246 | value: valued.get(&(day.clone(), workspace.clone(), bucket.clone())).copied().unwrap_or(0), |
| 247 | day, |
| 248 | workspace, |
| 249 | bucket, |
| 250 | }) |
| 251 | .collect(); |
| 252 | (days.into_values().collect(), workspaces) |
| 253 | } |
| 254 | |
| 255 | /// A month-end source's day, from the snapshots of what it had come to: |
| 256 | /// each day's figure less the day before's in the same month (the first |
| 257 | /// day of a month, or the first snapshot, is its own). |
| 258 | pub(crate) fn pending_deltas(snapshots: &[(String, String, String, i64, i64)]) -> Vec<UsageRow> { |
| 259 | // (day, workspace, source, cost, charge), any order. |
| 260 | let mut sorted = snapshots.to_vec(); |
| 261 | sorted.sort_by(|a, b| (&a.1, &a.2, &a.0).cmp(&(&b.1, &b.2, &b.0))); |
| 262 | let mut out = Vec::new(); |
| 263 | let mut previous: Option<&(String, String, String, i64, i64)> = None; |
| 264 | for snap in &sorted { |
| 265 | let (day, workspace, source, cost, charge) = snap; |
| 266 | let (before_cost, before_charge) = match previous { |
| 267 | Some(p) if p.1 == *workspace && p.2 == *source && p.0[..7] == day[..7] => (p.3, p.4), |
| 268 | _ => (0, 0), |
| 269 | }; |
| 270 | let (cost, charge) = ((cost - before_cost).max(0), (charge - before_charge).max(0)); |
| 271 | if cost > 0 || charge > 0 { |
| 272 | out.push(UsageRow { day: day.clone(), workspace: workspace.clone(), key: source.clone(), value: charge, cash: charge, cost }); |
| 273 | } |
| 274 | previous = Some(snap); |
| 275 | } |
| 276 | out |
| 277 | } |
| 278 | |
| 279 | /// Margin as a share of what was charged, in percent; None when nothing was. |
| 280 | pub(crate) fn margin_percent(revenue_micros: i64, cost_micros: i64) -> Option<f64> { |
| 281 | (revenue_micros > 0).then(|| (revenue_micros - cost_micros) as f64 * 100.0 / revenue_micros as f64) |
| 282 | } |
| 283 | |
| 284 | /// How far `ours` is from `theirs`, in percent of theirs; None when theirs |
| 285 | /// is nothing. |
| 286 | pub(crate) fn delta_percent(ours: f64, theirs: f64) -> Option<f64> { |
| 287 | (theirs > 0.0).then(|| (ours - theirs) * 100.0 / theirs) |
| 288 | } |
| 289 | |
| 290 | #[derive(Clone, Copy, Debug, PartialEq, Eq)] |
| 291 | pub(crate) enum DriftKind { |
| 292 | /// g1t counted a different number of units than Cloudflare did. |
| 293 | Count, |
| 294 | /// What Cloudflare charged differs from what the price book says the |
| 295 | /// same usage cost. |
| 296 | Cost, |
| 297 | /// Cloudflare charged for something nothing charges customers for. |
| 298 | Leak, |
| 299 | } |
| 300 | |
| 301 | impl DriftKind { |
| 302 | pub fn as_str(self) -> &'static str { |
| 303 | match self { |
| 304 | DriftKind::Count => "count", |
| 305 | DriftKind::Cost => "cost", |
| 306 | DriftKind::Leak => "leak", |
| 307 | } |
| 308 | } |
| 309 | } |
| 310 | |
| 311 | #[derive(Clone, Debug, PartialEq)] |
| 312 | pub(crate) struct Drift { |
| 313 | pub bucket: String, |
| 314 | pub kind: DriftKind, |
| 315 | pub ours: f64, |
| 316 | pub cloudflare: f64, |
| 317 | pub delta_percent: Option<f64>, |
| 318 | } |
| 319 | |
| 320 | /// Drift over a window for one bucket: counts more than `threshold` |
| 321 | /// percent apart, a bill that far from the price book's cost of the same |
| 322 | /// usage, and cost with nothing charged for it. Under `min_cost_micros` |
| 323 | /// in all, cost says nothing. |
| 324 | pub(crate) fn drifts(bucket: &str, days: &[ProductDay], threshold: f64, counted: bool, min_cost_micros: i64) -> Vec<Drift> { |
| 325 | let overhead = OVERHEAD.contains(&bucket); |
| 326 | let sum = |f: &dyn Fn(&ProductDay) -> f64| days.iter().map(f).sum::<f64>(); |
| 327 | let cf_cost = sum(&|d| d.cf_cost_micros as f64); |
| 328 | let own_cost = sum(&|d| d.own_cost_micros as f64); |
| 329 | let value = sum(&|d| d.value_micros as f64); |
| 330 | let (cf_quantity, own_quantity) = (sum(&|d| d.cf_quantity), sum(&|d| d.own_quantity)); |
| 331 | let mut out = Vec::new(); |
| 332 | if counted && cf_quantity > 0.0 { |
| 333 | let delta = delta_percent(own_quantity, cf_quantity); |
| 334 | if delta.is_some_and(|d| d.abs() > threshold) { |
| 335 | out.push(Drift { bucket: bucket.into(), kind: DriftKind::Count, ours: own_quantity, cloudflare: cf_quantity, delta_percent: delta }); |
| 336 | } |
| 337 | } |
| 338 | let enough = cf_cost.max(own_cost) >= min_cost_micros as f64; |
| 339 | if enough && !overhead && cf_cost > 0.0 && own_cost > 0.0 && !NOT_CLOUDFLARE.contains(&bucket) { |
| 340 | let delta = delta_percent(own_cost, cf_cost); |
| 341 | if delta.is_some_and(|d| d.abs() > threshold) { |
| 342 | out.push(Drift { bucket: bucket.into(), kind: DriftKind::Cost, ours: own_cost, cloudflare: cf_cost, delta_percent: delta }); |
| 343 | } |
| 344 | } |
| 345 | if !overhead && cf_cost >= min_cost_micros as f64 && value <= 0.0 { |
| 346 | out.push(Drift { bucket: bucket.into(), kind: DriftKind::Leak, ours: value, cloudflare: cf_cost, delta_percent: None }); |
| 347 | } |
| 348 | out |
| 349 | } |
| 350 | |
| 351 | /// When the last `days` in a row (each with enough cost to say something) |
| 352 | /// were all under the floor: the first of them and the worst margin. |
| 353 | /// Each item is a day's (day, revenue, cost). |
| 354 | pub(crate) fn breach(series: &[(String, i64, i64)], floor_percent: f64, days: usize, min_cost_micros: i64) -> Option<(String, f64)> { |
| 355 | if days == 0 || series.len() < days { |
| 356 | return None; |
| 357 | } |
| 358 | let tail = &series[series.len() - days..]; |
| 359 | let mut worst = f64::INFINITY; |
| 360 | for (_, revenue, cost) in tail { |
| 361 | if *cost < min_cost_micros { |
| 362 | return None; |
| 363 | } |
| 364 | let margin = margin_percent(*revenue, *cost).unwrap_or(-100.0); |
| 365 | if margin >= floor_percent { |
| 366 | return None; |
| 367 | } |
| 368 | worst = worst.min(margin); |
| 369 | } |
| 370 | Some((tail[0].0.clone(), worst)) |
| 371 | } |
| 372 | |
| 373 | /// `total` shared out in proportion to `weights`, in whole millionths that |
| 374 | /// add up to it exactly (largest remainder first). Nothing to share, or no |
| 375 | /// weight, shares nothing. |
| 376 | pub(crate) fn attribute(total: i64, weights: &[(String, f64)]) -> Vec<(String, i64)> { |
| 377 | let mut merged: BTreeMap<String, f64> = BTreeMap::new(); |
| 378 | for (key, w) in weights { |
| 379 | *merged.entry(key.clone()).or_default() += w.max(0.0); |
| 380 | } |
| 381 | let sum: f64 = merged.values().sum(); |
| 382 | if total <= 0 || sum <= 0.0 { |
| 383 | return Vec::new(); |
| 384 | } |
| 385 | let mut shares: Vec<(String, i64, f64)> = merged |
| 386 | .into_iter() |
| 387 | .map(|(key, w)| { |
| 388 | let exact = total as f64 * w / sum; |
| 389 | (key, exact.floor() as i64, exact - exact.floor()) |
| 390 | }) |
| 391 | .collect(); |
| 392 | let mut left = total - shares.iter().map(|s| s.1).sum::<i64>(); |
| 393 | let mut order: Vec<usize> = (0..shares.len()).collect(); |
| 394 | order.sort_by(|a, b| shares[*b].2.total_cmp(&shares[*a].2).then(shares[*a].0.cmp(&shares[*b].0))); |
| 395 | for index in order { |
| 396 | if left <= 0 { |
| 397 | break; |
| 398 | } |
| 399 | shares[index].1 += 1; |
| 400 | left -= 1; |
| 401 | } |
| 402 | shares.into_iter().filter(|s| s.1 > 0).map(|(key, micros, _)| (key, micros)).collect() |
| 403 | } |
| 404 | |
| 405 | /// Workspaces that cost g1t more than `factor` times what they paid, with |
| 406 | /// at least `floor_micros` of cost: each (workspace, cost, revenue), the |
| 407 | /// biggest gap first. |
| 408 | /// What the overall alert says: the money as money, and a percentage only |
| 409 | /// while there is enough coming in for one to mean something (a few cents |
| 410 | /// against dollars of cost reads as -8000%). |
| 411 | pub(crate) fn overall_detail(took: i64, spent: i64, days: usize, floor: f64, worst: f64) -> String { |
| 412 | if took < 1_000_000 * days as i64 { |
| 413 | return format!( |
| 414 | "All of g1t, comped workspaces left out: took in {} against {} of Cloudflare's bill over {days} days.", |
| 415 | dollars(took), |
| 416 | dollars(spent) |
| 417 | ); |
| 418 | } |
| 419 | format!("All of g1t, comped workspaces left out: money in against Cloudflare's bill under {floor:.0}% for {days} days running, as low as {worst:.1}%.") |
| 420 | } |
| 421 | |
| 422 | pub(crate) fn anomalies(rows: &[(String, i64, i64)], factor: f64, floor_micros: i64) -> Vec<(String, i64, i64)> { |
| 423 | let mut out: Vec<(String, i64, i64)> = rows |
| 424 | .iter() |
| 425 | .filter(|(_, cost, revenue)| *cost >= floor_micros && *cost as f64 > *revenue as f64 * factor) |
| 426 | .cloned() |
| 427 | .collect(); |
| 428 | out.sort_by(|a, b| (b.1 - b.2).cmp(&(a.1 - a.2)).then(a.0.cmp(&b.0))); |
| 429 | out |
| 430 | } |
| 431 | |
| 432 | /// Cloudflare's marginal rate for one of its units: the median over the |
| 433 | /// charged days of cost over quantity, in dollars. None while the included |
| 434 | /// amounts still cover it. Each item is a day's (quantity, cost). |
| 435 | pub(crate) fn billed_rate(days: &[(f64, f64)]) -> Option<f64> { |
| 436 | let mut rates: Vec<f64> = days.iter().filter(|(q, c)| *q > 0.0 && *c > 0.0).map(|(q, c)| c / q).collect(); |
| 437 | if rates.is_empty() { |
| 438 | return None; |
| 439 | } |
| 440 | rates.sort_by(f64::total_cmp); |
| 441 | Some(rates[rates.len() / 2]) |
| 442 | } |
| 443 | |
| 444 | /// What one of g1t's units costs, from Cloudflare's rate per its own unit |
| 445 | /// and how many of Cloudflare's units each of g1t's took: if Cloudflare |
| 446 | /// counts three operations for every git operation g1t counts, a git |
| 447 | /// operation costs three of Cloudflare's. None without enough of g1t's |
| 448 | /// units to say. |
| 449 | pub(crate) fn derived_unit_cost(rate_per_cf_unit: f64, cf_units: f64, own_units: f64) -> Option<f64> { |
| 450 | (own_units >= MIN_UNITS && cf_units > 0.0 && rate_per_cf_unit > 0.0).then(|| rate_per_cf_unit * cf_units / own_units) |
| 451 | } |
| 452 | |
| 453 | /// How many units a price is per: `1,000 operations` → 1,000, `million |
| 454 | /// requests` → 1,000,000, `second` → 1. |
| 455 | pub(crate) fn unit_size(unit: &str) -> f64 { |
| 456 | let first = unit.split_whitespace().next().unwrap_or_default().replace(',', ""); |
| 457 | match first.as_str() { |
| 458 | "million" => 1_000_000.0, |
| 459 | "thousand" => 1_000.0, |
| 460 | n => n.parse().unwrap_or(1.0), |
| 461 | } |
| 462 | } |
| 463 | |
| 464 | fn day_before(day: &str, days: u64) -> String { |
| 465 | let ms = g1t_contracts::time::parse_rfc3339(&format!("{day}T00:00:00Z")).unwrap_or(0); |
| 466 | rfc3339(ms.saturating_sub(days * DAY_MS))[..10].to_owned() |
| 467 | } |
| 468 | |
| 469 | /// Dollars to the cent from a dollar up, finer below: `$17.02`, `$0.063`. |
| 470 | fn dollars(micros: i64) -> String { |
| 471 | if micros.abs() >= 1_000_000 { |
| 472 | let cents = (micros as f64 / 10_000.0).round() as i64; |
| 473 | format!("{}${}.{:02}", if cents < 0 { "-" } else { "" }, cents.abs() / 100, cents.abs() % 100) |
| 474 | } else { |
| 475 | crate::features::dollars(micros) |
| 476 | } |
| 477 | } |
| 478 | |
| 479 | /// The days a plan payment is spread over. |
| 480 | const PLAN_DAYS: u64 = 30; |
| 481 | |
| 482 | /// `micros` paid on `day` spread evenly over `days` days from it, in |
| 483 | /// whole micros that add up to it (the first days take the remainder). |
| 484 | pub(crate) fn spread(day: &str, micros: i64, days: u64) -> Vec<(String, i64)> { |
| 485 | if micros <= 0 || days == 0 { |
| 486 | return Vec::new(); |
| 487 | } |
| 488 | let start = g1t_contracts::time::parse_rfc3339(&format!("{}T00:00:00Z", &day[..10.min(day.len())])).unwrap_or(0); |
| 489 | let each = micros / days as i64; |
| 490 | let rest = micros % days as i64; |
| 491 | (0..days) |
| 492 | .map(|n| (rfc3339(start + n * DAY_MS)[..10].to_owned(), each + i64::from((n as i64) < rest))) |
| 493 | .collect() |
| 494 | } |
| 495 | |
| 496 | // --------------------------------------------------------------------- |
| 497 | // The daily run, and what sudo reads. |
| 498 | // --------------------------------------------------------------------- |
| 499 | |
| 500 | #[derive(Serialize)] |
| 501 | struct Mail<'a> { |
| 502 | to: &'a str, |
| 503 | from: &'a str, |
| 504 | subject: &'a str, |
| 505 | text: String, |
| 506 | html: String, |
| 507 | } |
| 508 | |
| 509 | fn escape(text: &str) -> String { |
| 510 | text.replace('&', "&").replace('<', "<").replace('>', ">").replace('"', """) |
| 511 | } |
| 512 | |
| 513 | /// Emails staff through Cloudflare Email Sending, the `EMAIL` binding. |
| 514 | pub(crate) async fn email_staff(env: &Env, to: &str, subject: &str, lines: &[String]) -> Result<()> { |
| 515 | let link = "https://sudo.g1t.sh/costs"; |
| 516 | let text = format!("{}\n\nCosts & margin: {link}\n\nSent by g1t-billing's margin guard (COSTS_ALERT_EMAIL).\n", lines.join("\n\n")); |
| 517 | let mut html = String::from("<div style=\"font-family:system-ui,sans-serif;max-width:560px;margin:0 auto;padding:24px 16px;color:#16150f\">"); |
| 518 | for line in lines { |
| 519 | html.push_str(&format!("<p style=\"font-size:15px;line-height:1.6\">{}</p>", escape(line))); |
| 520 | } |
| 521 | html.push_str(&format!( |
| 522 | "<p><a href=\"{link}\">Open Costs & margin in sudo</a></p><p style=\"font-size:13px;color:#6e6a5e\">Sent by g1t-billing's margin guard (COSTS_ALERT_EMAIL).</p></div>" |
| 523 | )); |
| 524 | let mail = Mail { to, from: "g1t <noreply@g1t.sh>", subject, text, html }; |
| 525 | let binding = g1t_kit::js::binding(env, "EMAIL")?; |
| 526 | g1t_kit::js::call(&binding, "send", &[g1t_kit::js::to_js(&mail)?]).await?; |
| 527 | Ok(()) |
| 528 | } |
| 529 | |
| 530 | #[derive(Deserialize)] |
| 531 | struct AlertRow { |
| 532 | id: String, |
| 533 | kind: String, |
| 534 | subject: String, |
| 535 | detail: String, |
| 536 | since: String, |
| 537 | opened_at: String, |
| 538 | emailed_at: Option<String>, |
| 539 | } |
| 540 | |
| 541 | impl From<AlertRow> for MarginAlert { |
| 542 | fn from(r: AlertRow) -> Self { |
| 543 | MarginAlert { id: r.id, kind: r.kind, subject: r.subject, detail: r.detail, since: r.since, opened_at: r.opened_at, emailed_at: r.emailed_at } |
| 544 | } |
| 545 | } |
| 546 | |
| 547 | #[derive(Deserialize)] |
| 548 | struct MarginRow { |
| 549 | day: String, |
| 550 | bucket: String, |
| 551 | cf_cost_micros: i64, |
| 552 | own_cost_micros: i64, |
| 553 | value_micros: i64, |
| 554 | cash_micros: i64, |
| 555 | cf_quantity: f64, |
| 556 | own_quantity: f64, |
| 557 | } |
| 558 | |
| 559 | impl From<MarginRow> for ProductDay { |
| 560 | fn from(r: MarginRow) -> Self { |
| 561 | ProductDay { |
| 562 | day: r.day, |
| 563 | bucket: r.bucket, |
| 564 | cf_cost_micros: r.cf_cost_micros, |
| 565 | own_cost_micros: r.own_cost_micros, |
| 566 | value_micros: r.value_micros, |
| 567 | cash_micros: r.cash_micros, |
| 568 | cf_quantity: r.cf_quantity, |
| 569 | own_quantity: r.own_quantity, |
| 570 | } |
| 571 | } |
| 572 | } |
| 573 | |
| 574 | impl Billing { |
| 575 | /// The day's work: read Cloudflare's bill and g1t's own counts, |
| 576 | /// reconcile, look for drift, measure unit costs, apply prices whose |
| 577 | /// day has come, and raise or clear alerts. |
| 578 | pub(crate) async fn costs_daily(&self, env: &Env, keeper: &crate::keeper::Keeper) -> Result<CostsRun> { |
| 579 | let mut run = CostsRun::default(); |
| 580 | let (since, until) = match self.read_cloudflare(keeper, &mut run.problems).await? { |
| 581 | Some((since, until, lines)) => { |
| 582 | run.lines = lines; |
| 583 | (since, until) |
| 584 | } |
| 585 | // Without the bill, still reconcile what g1t knows itself, over |
| 586 | // the same days the bill would be read for. |
| 587 | None => { |
| 588 | #[derive(Deserialize)] |
| 589 | struct Last { |
| 590 | day: Option<String>, |
| 591 | } |
| 592 | let last = self.db.prepare("SELECT MAX(day) AS day FROM margin_days").first::<Last>(None).await?.and_then(|l| l.day); |
| 593 | costs::window(last.as_deref(), now_ms()) |
| 594 | } |
| 595 | }; |
| 596 | if let Err(error) = self.count_own(&since, &until).await { |
| 597 | run.problems.push(format!("g1t's own counts could not be read: {error}")); |
| 598 | } |
| 599 | self.snapshot_pending(&until).await?; |
| 600 | run.days = self.reconcile_range(&since, &until).await?; |
| 601 | let drift = self.find_drift(&until).await?; |
| 602 | run.proposals = self.measure_units(&until).await?; |
| 603 | self.apply_due_versions().await?; |
| 604 | run.alerts = self.raise_alerts(env, &until, &drift).await?; |
| 605 | if let Some(identity) = &self.identity |
| 606 | && let Err(error) = self.tell_owners_of_rises(identity).await |
| 607 | { |
| 608 | run.problems.push(format!("owners could not be told of a price rise: {error}")); |
| 609 | } |
| 610 | for problem in &run.problems { |
| 611 | worker::console_warn!("costs: {problem}"); |
| 612 | } |
| 613 | Ok(run) |
| 614 | } |
| 615 | |
| 616 | /// What each month-end source had come to by the end of `day`. |
| 617 | async fn snapshot_pending(&self, day: &str) -> Result<()> { |
| 618 | self.db |
| 619 | .prepare( |
| 620 | "INSERT INTO pending_days (day, workspace, source, cost_micros, charge_micros) |
| 621 | SELECT ?1, workspace, source, COALESCE(cost_micros, 0), COALESCE(charge_micros, 0) FROM pending_usage WHERE month = ?2 |
| 622 | ON CONFLICT (day, workspace, source) DO UPDATE SET cost_micros = excluded.cost_micros, charge_micros = excluded.charge_micros", |
| 623 | ) |
| 624 | .bind(&[day.into(), day[..7].into()])? |
| 625 | .run() |
| 626 | .await?; |
| 627 | Ok(()) |
| 628 | } |
| 629 | |
| 630 | /// What customers were charged on the days, by workspace and key. |
| 631 | async fn usage_rows(&self, since: &str, until: &str) -> Result<Vec<UsageRow>> { |
| 632 | #[derive(Deserialize)] |
| 633 | struct Row { |
| 634 | day: String, |
| 635 | workspace: String, |
| 636 | key: String, |
| 637 | internal: i64, |
| 638 | own_provider: i64, |
| 639 | cash: Option<i64>, |
| 640 | drawn: Option<i64>, |
| 641 | cost: Option<i64>, |
| 642 | } |
| 643 | let charged_here = crate::storage::CHARGED_HERE.iter().map(|s| format!("'{s}'")).collect::<Vec<_>>().join(", "); |
| 644 | let end = format!("{until}T23:59:59.999Z"); |
| 645 | let rows = self |
| 646 | .db |
| 647 | .prepare(format!( |
| 648 | "SELECT substr(created_at, 1, 10) AS day, workspace, |
| 649 | CASE WHEN task = 'deployments' AND reference LIKE 'deploy/%' THEN 'builds' ELSE COALESCE(task, 'other') END AS key, |
| 650 | CASE WHEN workspace IN ({internal}) THEN 1 ELSE 0 END AS internal, |
| 651 | CASE WHEN billed_to = 'workspace' THEN 1 ELSE 0 END AS own_provider, |
| 652 | -SUM(amount_micros) AS cash, |
| 653 | SUM(COALESCE(credit_micros, 0) + COALESCE(trial_micros, 0) + COALESCE(oss_micros, 0) + COALESCE(given_micros, 0)) AS drawn, |
| 654 | SUM(COALESCE(cost_micros, 0)) AS cost |
| 655 | FROM ledger |
| 656 | WHERE kind = 'usage' AND created_at >= ?1 AND created_at <= ?2 AND COALESCE(task, '') NOT IN ({charged_here}) |
| 657 | GROUP BY 1, 2, 3, 4, 5", |
| 658 | internal = crate::sales::INTERNAL_SQL |
| 659 | )) |
| 660 | .bind(&[since.into(), end.as_str().into()])? |
| 661 | .all() |
| 662 | .await? |
| 663 | .results::<Row>()?; |
| 664 | let mut out: Vec<UsageRow> = rows |
| 665 | .into_iter() |
| 666 | .map(|r| { |
| 667 | // A workspace's own model provider was paid there: no cost |
| 668 | // to g1t. g1t's own workspaces are valued at price. |
| 669 | let cost = if r.own_provider == 1 { 0 } else { r.cost.unwrap_or(0) }; |
| 670 | let cash = r.cash.unwrap_or(0); |
| 671 | let value = if r.internal == 1 { crate::credits::with_margin(cost, self.margin_percent) } else { cash + r.drawn.unwrap_or(0) }; |
| 672 | UsageRow { day: r.day, workspace: r.workspace, key: r.key, value, cash, cost } |
| 673 | }) |
| 674 | .collect(); |
| 675 | // Month-end sources, from their daily snapshots. |
| 676 | #[derive(Deserialize)] |
| 677 | struct Snap { |
| 678 | day: String, |
| 679 | workspace: String, |
| 680 | source: String, |
| 681 | cost_micros: i64, |
| 682 | charge_micros: i64, |
| 683 | } |
| 684 | let snaps = self |
| 685 | .db |
| 686 | .prepare("SELECT day, workspace, source, cost_micros, charge_micros FROM pending_days WHERE day >= ?1 AND day <= ?2") |
| 687 | .bind(&[day_before(since, 1).into(), until.into()])? |
| 688 | .all() |
| 689 | .await? |
| 690 | .results::<Snap>()? |
| 691 | .into_iter() |
| 692 | .filter(|s| crate::storage::CHARGED_HERE.contains(&s.source.as_str()) || s.source == "domains") |
| 693 | .map(|s| (s.day, s.workspace, s.source, s.cost_micros, s.charge_micros)) |
| 694 | .collect::<Vec<_>>(); |
| 695 | out.extend(pending_deltas(&snaps).into_iter().filter(|u| u.day.as_str() >= since)); |
| 696 | // The plan's price, spread over the 30 days it pays for, so a month's |
| 697 | // payment does not read as one very good day and 29 bad ones. |
| 698 | #[derive(Deserialize)] |
| 699 | struct Plan { |
| 700 | day: String, |
| 701 | workspace: String, |
| 702 | micros: Option<i64>, |
| 703 | } |
| 704 | let plans = self |
| 705 | .db |
| 706 | .prepare( |
| 707 | "SELECT substr(paid_at, 1, 10) AS day, workspace, SUM(amount_micros) AS micros FROM plan_payments |
| 708 | WHERE paid_at >= ?1 AND paid_at <= ?2 GROUP BY 1, 2", |
| 709 | ) |
| 710 | .bind(&[day_before(since, PLAN_DAYS - 1).into(), end.as_str().into()])? |
| 711 | .all() |
| 712 | .await? |
| 713 | .results::<Plan>()?; |
| 714 | for p in plans { |
| 715 | for (day, micros) in spread(&p.day, p.micros.unwrap_or(0), PLAN_DAYS) { |
| 716 | if day.as_str() >= since && day.as_str() <= until { |
| 717 | out.push(UsageRow { day, workspace: p.workspace.clone(), key: "plan".into(), value: micros, cash: micros, cost: 0 }); |
| 718 | } |
| 719 | } |
| 720 | } |
| 721 | Ok(out) |
| 722 | } |
| 723 | |
| 724 | /// Reconciles the days and writes `margin_days` and `workspace_costs`. |
| 725 | async fn reconcile_range(&self, since: &str, until: &str) -> Result<u32> { |
| 726 | let rules = self.rules().await?; |
| 727 | #[derive(Deserialize)] |
| 728 | struct Map { |
| 729 | key: String, |
| 730 | bucket: String, |
| 731 | } |
| 732 | let revenue_map: BTreeMap<String, String> = self |
| 733 | .db |
| 734 | .prepare("SELECT key, bucket FROM revenue_map") |
| 735 | .all() |
| 736 | .await? |
| 737 | .results::<Map>()? |
| 738 | .into_iter() |
| 739 | .map(|m| (m.key, m.bucket)) |
| 740 | .collect(); |
| 741 | let lines = self |
| 742 | .db |
| 743 | .prepare("SELECT day, source, product, meter, quantity, cost_usd FROM cost_lines WHERE day >= ?1 AND day <= ?2") |
| 744 | .bind(&[since.into(), until.into()])? |
| 745 | .all() |
| 746 | .await? |
| 747 | .results::<LineRow>()?; |
| 748 | let own = self |
| 749 | .db |
| 750 | .prepare("SELECT day, meter, workspace, quantity FROM own_counts WHERE day >= ?1 AND day <= ?2") |
| 751 | .bind(&[since.into(), until.into()])? |
| 752 | .all() |
| 753 | .await? |
| 754 | .results::<OwnRow>()?; |
| 755 | let usage = self.usage_rows(since, until).await?; |
| 756 | let (days, workspaces) = fold(&rules, &revenue_map, &lines, &own, &usage); |
| 757 | let now = rfc3339(now_ms()); |
| 758 | self.db |
| 759 | .batch(vec![ |
| 760 | self.db.prepare("DELETE FROM margin_days WHERE day >= ?1 AND day <= ?2").bind(&[since.into(), until.into()])?, |
| 761 | self.db.prepare("DELETE FROM workspace_costs WHERE day >= ?1 AND day <= ?2").bind(&[since.into(), until.into()])?, |
| 762 | ]) |
| 763 | .await?; |
| 764 | for chunk in days.chunks(50) { |
| 765 | let mut statements = Vec::with_capacity(chunk.len()); |
| 766 | for d in chunk { |
| 767 | statements.push( |
| 768 | self.db |
| 769 | .prepare( |
| 770 | "INSERT OR REPLACE INTO margin_days (day, bucket, cf_cost_micros, own_cost_micros, value_micros, cash_micros, cf_quantity, own_quantity, computed_at) |
| 771 | VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)", |
| 772 | ) |
| 773 | .bind(&[ |
| 774 | d.day.as_str().into(), |
| 775 | d.bucket.as_str().into(), |
| 776 | (d.cf_cost_micros as f64).into(), |
| 777 | (d.own_cost_micros as f64).into(), |
| 778 | (d.value_micros as f64).into(), |
| 779 | (d.cash_micros as f64).into(), |
| 780 | d.cf_quantity.into(), |
| 781 | d.own_quantity.into(), |
| 782 | now.as_str().into(), |
| 783 | ])?, |
| 784 | ); |
| 785 | } |
| 786 | self.db.batch(statements).await?; |
| 787 | } |
| 788 | for chunk in workspaces.chunks(50) { |
| 789 | let mut statements = Vec::with_capacity(chunk.len()); |
| 790 | for w in chunk { |
| 791 | statements.push( |
| 792 | self.db |
| 793 | .prepare("INSERT OR REPLACE INTO workspace_costs (day, workspace, bucket, cost_micros, revenue_micros, value_micros) VALUES (?, ?, ?, ?, ?, ?)") |
| 794 | .bind(&[ |
| 795 | w.day.as_str().into(), |
| 796 | w.workspace.as_str().into(), |
| 797 | w.bucket.as_str().into(), |
| 798 | (w.cost as f64).into(), |
| 799 | (w.revenue as f64).into(), |
| 800 | (w.value as f64).into(), |
| 801 | ])?, |
| 802 | ); |
| 803 | } |
| 804 | self.db.batch(statements).await?; |
| 805 | } |
| 806 | Ok(costs::days_between(since, until).len() as u32) |
| 807 | } |
| 808 | |
| 809 | async fn margin_days(&self, since: &str, until: &str) -> Result<Vec<ProductDay>> { |
| 810 | Ok(self |
| 811 | .db |
| 812 | .prepare("SELECT * FROM margin_days WHERE day >= ?1 AND day <= ?2 ORDER BY day, bucket") |
| 813 | .bind(&[since.into(), until.into()])? |
| 814 | .all() |
| 815 | .await? |
| 816 | .results::<MarginRow>()? |
| 817 | .into_iter() |
| 818 | .map(ProductDay::from) |
| 819 | .collect()) |
| 820 | } |
| 821 | |
| 822 | /// Drift over the last week, written to `cost_drift` (replacing the |
| 823 | /// last run's), with unmapped Cloudflare meters as leaks. |
| 824 | async fn find_drift(&self, until: &str) -> Result<Vec<(Drift, String)>> { |
| 825 | let since = day_before(until, DRIFT_DAYS - 1); |
| 826 | let settings = self.cost_settings().await?; |
| 827 | let rules = self.rules().await?; |
| 828 | let days = self.margin_days(&since, until).await?; |
| 829 | let mut by: BTreeMap<String, Vec<ProductDay>> = BTreeMap::new(); |
| 830 | for d in days { |
| 831 | by.entry(d.bucket.clone()).or_default().push(d); |
| 832 | } |
| 833 | let mut found = Vec::new(); |
| 834 | for (bucket, days) in &by { |
| 835 | let bucket_rules: Vec<&Rule> = rules.iter().filter(|r| &r.bucket == bucket).collect(); |
| 836 | let threshold = bucket_rules.iter().map(|r| r.drift_percent).fold(f64::INFINITY, f64::min); |
| 837 | let threshold = if threshold.is_finite() { threshold } else { 10.0 }; |
| 838 | let counted = bucket_rules.iter().any(|r| r.own_meter.is_some()); |
| 839 | for drift in drifts(bucket, days, threshold, counted, settings.min_daily_cost_micros) { |
| 840 | let title = costs::bucket_title(bucket); |
| 841 | let detail = match drift.kind { |
| 842 | DriftKind::Count => format!( |
| 843 | "{title}: g1t counted {}, Cloudflare {} over the last {DRIFT_DAYS} days ({:+.1}%). Customers are charged for what g1t counts; check what Cloudflare counts as a unit and change the repos service's operation_mapping (set_operation_mapping).", |
| 844 | crate::features::thousands(drift.ours.max(0.0).round() as u64), |
| 845 | crate::features::thousands(drift.cloudflare.max(0.0).round() as u64), |
| 846 | drift.delta_percent.unwrap_or(0.0) |
| 847 | ), |
| 848 | DriftKind::Cost => format!( |
| 849 | "{title}: Cloudflare charged {} over the last {DRIFT_DAYS} days; the price book's cost of the same usage is {} ({:+.1}%). A price may be stale: see the proposals.", |
| 850 | dollars(drift.cloudflare as i64), |
| 851 | dollars(drift.ours as i64), |
| 852 | drift.delta_percent.unwrap_or(0.0) |
| 853 | ), |
| 854 | DriftKind::Leak if bucket == UNMAPPED => { |
| 855 | format!("Cloudflare charged {} for meters no mapping claims. Map them on Costs & margin.", dollars(drift.cloudflare as i64)) |
| 856 | } |
| 857 | DriftKind::Leak => format!( |
| 858 | "{title}: Cloudflare charged {} over the last {DRIFT_DAYS} days and customers were charged nothing for it.", |
| 859 | dollars(drift.cloudflare as i64) |
| 860 | ), |
| 861 | }; |
| 862 | found.push((drift, detail)); |
| 863 | } |
| 864 | } |
| 865 | let now = rfc3339(now_ms()); |
| 866 | let mut statements = vec![self.db.prepare("DELETE FROM cost_drift")]; |
| 867 | for (drift, detail) in &found { |
| 868 | statements.push( |
| 869 | self.db |
| 870 | .prepare("INSERT OR REPLACE INTO cost_drift (bucket, kind, ours, cloudflare, delta_percent, detail, found_at) VALUES (?, ?, ?, ?, ?, ?, ?)") |
| 871 | .bind(&[ |
| 872 | drift.bucket.as_str().into(), |
| 873 | drift.kind.as_str().into(), |
| 874 | drift.ours.into(), |
| 875 | drift.cloudflare.into(), |
| 876 | drift.delta_percent.map_or(JsValue::NULL, JsValue::from), |
| 877 | detail.as_str().into(), |
| 878 | now.as_str().into(), |
| 879 | ])?, |
| 880 | ); |
| 881 | } |
| 882 | self.db.batch(statements).await?; |
| 883 | Ok(found) |
| 884 | } |
| 885 | |
| 886 | /// Unit costs from the bill for mappings that scale to g1t's own count |
| 887 | /// (git operations), proposed to the price book. |
| 888 | async fn measure_units(&self, until: &str) -> Result<u32> { |
| 889 | #[derive(Deserialize)] |
| 890 | struct Scaled { |
| 891 | product: String, |
| 892 | meter: String, |
| 893 | price_meter: String, |
| 894 | own_meter: String, |
| 895 | unit: Option<String>, |
| 896 | } |
| 897 | let scaled = self |
| 898 | .db |
| 899 | .prepare( |
| 900 | "SELECT m.product, m.meter, m.price_meter, m.own_meter, p.unit FROM cost_map m LEFT JOIN prices p ON p.meter = m.price_meter |
| 901 | WHERE m.scale_to_own = 1 AND m.price_meter IS NOT NULL AND m.own_meter IS NOT NULL", |
| 902 | ) |
| 903 | .all() |
| 904 | .await? |
| 905 | .results::<Scaled>()?; |
| 906 | let since = day_before(until, MEASURE_DAYS - 1); |
| 907 | let rules = self.rules().await?; |
| 908 | let mut proposed = 0; |
| 909 | for s in scaled { |
| 910 | #[derive(Deserialize)] |
| 911 | struct Day { |
| 912 | product: String, |
| 913 | meter: String, |
| 914 | quantity: f64, |
| 915 | cost_usd: f64, |
| 916 | } |
| 917 | let lines = self |
| 918 | .db |
| 919 | .prepare("SELECT product, meter, quantity, cost_usd FROM cost_lines WHERE source = ?1 AND product = ?2 AND day >= ?3 AND day <= ?4") |
| 920 | .bind(&[SOURCE_BILLABLE.into(), s.product.as_str().into(), since.as_str().into(), until.into()])? |
| 921 | .all() |
| 922 | .await? |
| 923 | .results::<Day>()?; |
| 924 | // Only the lines this very mapping claims. |
| 925 | let mine: Vec<(f64, f64)> = lines |
| 926 | .iter() |
| 927 | .filter(|l| costs::classify(&rules, &l.product, &l.meter).is_some_and(|r| r.product == s.product && r.meter == s.meter)) |
| 928 | .map(|l| (l.quantity, l.cost_usd)) |
| 929 | .collect(); |
| 930 | let Some(rate) = billed_rate(&mine) else { continue }; |
| 931 | let cf_units: f64 = mine.iter().map(|(q, _)| q).sum(); |
| 932 | #[derive(Deserialize)] |
| 933 | struct Own { |
| 934 | total: Option<f64>, |
| 935 | } |
| 936 | let own_units = self |
| 937 | .db |
| 938 | .prepare("SELECT SUM(quantity) AS total FROM own_counts WHERE meter = ?1 AND day >= ?2 AND day <= ?3") |
| 939 | .bind(&[s.own_meter.as_str().into(), since.as_str().into(), until.into()])? |
| 940 | .first::<Own>(None) |
| 941 | .await? |
| 942 | .and_then(|o| o.total) |
| 943 | .unwrap_or(0.0); |
| 944 | let Some(per_unit) = derived_unit_cost(rate, cf_units, own_units) else { continue }; |
| 945 | let size = unit_size(s.unit.as_deref().unwrap_or("1")); |
| 946 | let measured = per_unit * size * 1_000_000.0; |
| 947 | let reason = format!( |
| 948 | "Cloudflare billed ${:.4} per 1,000 of its units and counted {:.2} of them for each one g1t counted over the last {MEASURE_DAYS} days ({} against {})", |
| 949 | rate * 1000.0, |
| 950 | cf_units / own_units, |
| 951 | crate::features::thousands(cf_units.round() as u64), |
| 952 | crate::features::thousands(own_units.round() as u64) |
| 953 | ); |
| 954 | if self.propose(&s.price_meter, measured, &reason, "reconciler").await?.is_some() { |
| 955 | proposed += 1; |
| 956 | } |
| 957 | } |
| 958 | Ok(proposed) |
| 959 | } |
| 960 | |
| 961 | /// Opens, updates and closes margin alerts, and emails staff about new |
| 962 | /// ones (and open ones each week). |
| 963 | async fn raise_alerts(&self, env: &Env, until: &str, drift: &[(Drift, String)]) -> Result<u32> { |
| 964 | let settings = self.cost_settings().await?; |
| 965 | let since = day_before(until, u64::from(settings.alert_days.max(1)) - 1); |
| 966 | let days = self.margin_days(&since, until).await?; |
| 967 | let mut conditions: Vec<(String, String, String, String)> = Vec::new(); |
| 968 | // Each product under the floor. |
| 969 | let mut by: BTreeMap<String, Vec<(String, i64, i64)>> = BTreeMap::new(); |
| 970 | let mut all: BTreeMap<String, (i64, i64)> = BTreeMap::new(); |
| 971 | for d in &days { |
| 972 | let overall = all.entry(d.day.clone()).or_default(); |
| 973 | overall.0 += d.cash_micros; |
| 974 | overall.1 += d.cost(); |
| 975 | if !OVERHEAD.contains(&d.bucket.as_str()) && d.bucket != UNMAPPED { |
| 976 | by.entry(d.bucket.clone()).or_default().push((d.day.clone(), d.value_micros, d.cost())); |
| 977 | } |
| 978 | } |
| 979 | let floor = settings.margin_floor_percent; |
| 980 | let n = settings.alert_days as usize; |
| 981 | for (bucket, series) in &by { |
| 982 | if let Some((from, worst)) = breach(series, floor, n, settings.min_daily_cost_micros) { |
| 983 | conditions.push(( |
| 984 | "margin".into(), |
| 985 | bucket.clone(), |
| 986 | format!("{}: margin under {floor:.0}% for {n} days running, as low as {worst:.1}%.", costs::bucket_title(bucket)), |
| 987 | from, |
| 988 | )); |
| 989 | } |
| 990 | } |
| 991 | // Comped workspaces' share is a budget g1t chose to spend, watched |
| 992 | // on its own (budget.rs): not part of whether what is sold pays. |
| 993 | for (day, cost) in self.comped_costs(&since, until).await? { |
| 994 | if let Some(overall) = all.get_mut(&day) { |
| 995 | overall.1 = (overall.1 - cost).max(0); |
| 996 | } |
| 997 | } |
| 998 | let series: Vec<(String, i64, i64)> = all.into_iter().map(|(day, (revenue, cost))| (day, revenue, cost)).collect(); |
| 999 | if let Some((from, worst)) = breach(&series, floor, n, settings.min_daily_cost_micros) { |
| 1000 | let tail = &series[series.len().saturating_sub(n)..]; |
| 1001 | let (took, spent) = tail.iter().fold((0i64, 0i64), |(r, c), (_, revenue, cost)| (r + revenue, c + cost)); |
| 1002 | conditions.push(("overall".into(), "g1t".into(), overall_detail(took, spent, n, floor, worst), from)); |
| 1003 | } |
| 1004 | for (d, detail) in drift { |
| 1005 | let kind = if d.kind == DriftKind::Leak { "leak" } else { "drift" }; |
| 1006 | conditions.push((kind.into(), format!("{}:{}", d.bucket, d.kind.as_str()), detail.clone(), until.to_owned())); |
| 1007 | } |
| 1008 | // Workspaces costing more than they pay. |
| 1009 | for (workspace, cost, revenue) in self.workspace_anomalies(until, &settings).await? { |
| 1010 | conditions.push(( |
| 1011 | "workspace".into(), |
| 1012 | workspace.clone(), |
| 1013 | format!( |
| 1014 | "{workspace} cost g1t {} on Cloudflare over {ANOMALY_DAYS} days, and its usage was priced at {}: its prices are below cost.", |
| 1015 | dollars(cost), |
| 1016 | dollars(revenue) |
| 1017 | ), |
| 1018 | day_before(until, ANOMALY_DAYS - 1), |
| 1019 | )); |
| 1020 | } |
| 1021 | |
| 1022 | let open = self |
| 1023 | .db |
| 1024 | .prepare("SELECT * FROM margin_alerts WHERE resolved_at IS NULL") |
| 1025 | .all() |
| 1026 | .await? |
| 1027 | .results::<AlertRow>()?; |
| 1028 | let now = now_ms(); |
| 1029 | let stamp = rfc3339(now); |
| 1030 | let mut to_email: Vec<String> = Vec::new(); |
| 1031 | let mut kept: BTreeSet<String> = BTreeSet::new(); |
| 1032 | for (kind, subject, detail, from) in &conditions { |
| 1033 | match open.iter().find(|a| &a.kind == kind && &a.subject == subject) { |
| 1034 | Some(alert) => { |
| 1035 | kept.insert(alert.id.clone()); |
| 1036 | self.db |
| 1037 | .prepare("UPDATE margin_alerts SET detail = ? WHERE id = ?") |
| 1038 | .bind(&[detail.as_str().into(), alert.id.as_str().into()])? |
| 1039 | .run() |
| 1040 | .await?; |
| 1041 | let stale = alert |
| 1042 | .emailed_at |
| 1043 | .as_deref() |
| 1044 | .and_then(g1t_contracts::time::parse_rfc3339) |
| 1045 | .is_none_or(|at| now.saturating_sub(at) >= REMIND_MS); |
| 1046 | if stale && kind != "workspace" { |
| 1047 | to_email.push(format!("Still open: {detail}")); |
| 1048 | kept.insert(format!("email:{}", alert.id)); |
| 1049 | } |
| 1050 | } |
| 1051 | None => { |
| 1052 | let id = new_id("mal", now); |
| 1053 | self.db |
| 1054 | .prepare("INSERT INTO margin_alerts (id, kind, subject, detail, since, opened_at) VALUES (?, ?, ?, ?, ?, ?)") |
| 1055 | .bind(&[id.as_str().into(), kind.as_str().into(), subject.as_str().into(), detail.as_str().into(), from.as_str().into(), stamp.as_str().into()])? |
| 1056 | .run() |
| 1057 | .await?; |
| 1058 | kept.insert(id.clone()); |
| 1059 | // A workspace's is for Reach out, not the inbox. |
| 1060 | if kind != "workspace" { |
| 1061 | to_email.push(detail.clone()); |
| 1062 | } |
| 1063 | kept.insert(format!("email:{id}")); |
| 1064 | } |
| 1065 | } |
| 1066 | } |
| 1067 | for alert in &open { |
| 1068 | if !kept.contains(&alert.id) { |
| 1069 | self.db |
| 1070 | .prepare("UPDATE margin_alerts SET resolved_at = ? WHERE id = ?") |
| 1071 | .bind(&[stamp.as_str().into(), alert.id.as_str().into()])? |
| 1072 | .run() |
| 1073 | .await?; |
| 1074 | } |
| 1075 | } |
| 1076 | let to = env.var("COSTS_ALERT_EMAIL").map(|v| v.to_string()).unwrap_or_default(); |
| 1077 | if !to_email.is_empty() && !to.trim().is_empty() { |
| 1078 | let subject = format!("[g1t costs] {} margin alert{}", to_email.len(), if to_email.len() == 1 { "" } else { "s" }); |
| 1079 | match email_staff(env, to.trim(), &subject, &to_email).await { |
| 1080 | Ok(()) => { |
| 1081 | for marker in kept.iter().filter_map(|k| k.strip_prefix("email:")) { |
| 1082 | self.db |
| 1083 | .prepare("UPDATE margin_alerts SET emailed_at = ? WHERE id = ?") |
| 1084 | .bind(&[stamp.as_str().into(), marker.into()])? |
| 1085 | .run() |
| 1086 | .await?; |
| 1087 | } |
| 1088 | } |
| 1089 | Err(error) => worker::console_error!("could not email the margin alerts: {error}"), |
| 1090 | } |
| 1091 | } |
| 1092 | Ok(conditions.len() as u32) |
| 1093 | } |
| 1094 | |
| 1095 | /// Workspaces costing g1t more than they pay over 30 days, not g1t's own. |
| 1096 | /// Each day's cost shared out to comped workspaces. |
| 1097 | async fn comped_costs(&self, since: &str, until: &str) -> Result<Vec<(String, i64)>> { |
| 1098 | #[derive(Deserialize)] |
| 1099 | struct Row { |
| 1100 | day: String, |
| 1101 | cost: Option<i64>, |
| 1102 | } |
| 1103 | Ok(self |
| 1104 | .db |
| 1105 | .prepare(format!( |
| 1106 | "SELECT day, SUM(cost_micros) AS cost FROM workspace_costs |
| 1107 | WHERE day >= ?1 AND day <= ?2 AND workspace IN ({}) GROUP BY day", |
| 1108 | crate::sales::INTERNAL_SQL |
| 1109 | )) |
| 1110 | .bind(&[since.into(), until.into()])? |
| 1111 | .all() |
| 1112 | .await? |
| 1113 | .results::<Row>()? |
| 1114 | .into_iter() |
| 1115 | .map(|r| (r.day, r.cost.unwrap_or(0))) |
| 1116 | .collect()) |
| 1117 | } |
| 1118 | |
| 1119 | async fn workspace_anomalies(&self, until: &str, settings: &CostSettings) -> Result<Vec<(String, i64, i64)>> { |
| 1120 | #[derive(Deserialize)] |
| 1121 | struct Row { |
| 1122 | workspace: String, |
| 1123 | cost: Option<i64>, |
| 1124 | revenue: Option<i64>, |
| 1125 | } |
| 1126 | let rows = self |
| 1127 | .db |
| 1128 | .prepare(format!( |
| 1129 | // Against what its usage was priced at, not the cash it |
| 1130 | // paid: a trial or a gift paying for usage is not a price |
| 1131 | // below cost. |
| 1132 | // Days from before value_micros was kept have none: only days |
| 1133 | // since the first one that does are compared. |
| 1134 | "SELECT workspace, SUM(cost_micros) AS cost, SUM(value_micros) AS revenue FROM workspace_costs |
| 1135 | WHERE day >= ?1 AND day <= ?2 AND workspace NOT IN ({}) |
| 1136 | AND day >= (SELECT MIN(day) FROM workspace_costs WHERE value_micros > 0) |
| 1137 | GROUP BY workspace", |
| 1138 | crate::sales::INTERNAL_SQL |
| 1139 | )) |
| 1140 | .bind(&[day_before(until, ANOMALY_DAYS - 1).into(), until.into()])? |
| 1141 | .all() |
| 1142 | .await? |
| 1143 | .results::<Row>()?; |
| 1144 | let rows: Vec<(String, i64, i64)> = rows.into_iter().map(|r| (r.workspace, r.cost.unwrap_or(0), r.revenue.unwrap_or(0))).collect(); |
| 1145 | Ok(anomalies(&rows, settings.anomaly_factor, settings.anomaly_floor_micros)) |
| 1146 | } |
| 1147 | |
| 1148 | /// For Reach out: workspaces with an open cost-over-revenue alert, |
| 1149 | /// each with its detail and cost. |
| 1150 | pub(crate) async fn costing_more_than_they_pay(&self) -> Result<Vec<(String, String, i64)>> { |
| 1151 | let alerts = self |
| 1152 | .db |
| 1153 | .prepare("SELECT * FROM margin_alerts WHERE resolved_at IS NULL AND kind = 'workspace' ORDER BY opened_at DESC LIMIT 50") |
| 1154 | .all() |
| 1155 | .await? |
| 1156 | .results::<AlertRow>()?; |
| 1157 | let mut out = Vec::new(); |
| 1158 | for alert in alerts { |
| 1159 | #[derive(Deserialize)] |
| 1160 | struct Cost { |
| 1161 | cost: Option<i64>, |
| 1162 | } |
| 1163 | let cost = self |
| 1164 | .db |
| 1165 | .prepare("SELECT SUM(cost_micros) AS cost FROM workspace_costs WHERE workspace = ? AND day >= ?") |
| 1166 | .bind(&[alert.subject.as_str().into(), alert.since.as_str().into()])? |
| 1167 | .first::<Cost>(None) |
| 1168 | .await? |
| 1169 | .and_then(|c| c.cost) |
| 1170 | .unwrap_or(0); |
| 1171 | out.push((alert.subject, alert.detail, cost)); |
| 1172 | } |
| 1173 | Ok(out) |
| 1174 | } |
| 1175 | |
| 1176 | /// `admin_cost_alerts`: what sudo's banner says. |
| 1177 | pub(crate) async fn admin_cost_alerts(&self, _: AdminCostAlertsArgs) -> Result<Vec<MarginAlert>> { |
| 1178 | Ok(self |
| 1179 | .db |
| 1180 | .prepare("SELECT * FROM margin_alerts WHERE resolved_at IS NULL ORDER BY opened_at DESC LIMIT 50") |
| 1181 | .all() |
| 1182 | .await? |
| 1183 | .results::<AlertRow>()? |
| 1184 | .into_iter() |
| 1185 | .map(MarginAlert::from) |
| 1186 | .collect()) |
| 1187 | } |
| 1188 | |
| 1189 | /// `admin_run_costs`: the daily run, now. |
| 1190 | pub(crate) async fn admin_run_costs(&self, env: &Env, a: AdminRunCostsArgs) -> Result<Outcome<CostsRun>> { |
| 1191 | let keeper = crate::keeper::Keeper::from_env(env); |
| 1192 | let run = self.costs_daily(env, &keeper).await?; |
| 1193 | if !a.by.is_empty() { |
| 1194 | self.audit( |
| 1195 | "costs", |
| 1196 | "costs_run", |
| 1197 | &format!("{} lines, {} days, {} proposals, {} alerts", run.lines, run.days, run.proposals, run.alerts), |
| 1198 | &a.by, |
| 1199 | ) |
| 1200 | .await?; |
| 1201 | } |
| 1202 | Ok(Outcome::Ok(run)) |
| 1203 | } |
| 1204 | |
| 1205 | /// `admin_set_cost_mapping`. |
| 1206 | pub(crate) async fn admin_set_cost_mapping(&self, a: AdminSetCostMappingArgs) -> Result<Outcome<CostMapping>> { |
| 1207 | let product = costs::slug(&a.product); |
| 1208 | let meter = if a.meter.trim() == "*" { "*".to_owned() } else { costs::slug(&a.meter) }; |
| 1209 | if product.is_empty() || meter.is_empty() { |
| 1210 | return Ok(Outcome::fail(FailureCode::Invalid, "Name Cloudflare's product and a meter (or * for all of it).")); |
| 1211 | } |
| 1212 | let now = rfc3339(now_ms()); |
| 1213 | if a.remove { |
| 1214 | self.db |
| 1215 | .prepare("DELETE FROM cost_map WHERE product = ? AND meter = ?") |
| 1216 | .bind(&[product.as_str().into(), meter.as_str().into()])? |
| 1217 | .run() |
| 1218 | .await?; |
| 1219 | self.audit("costs", "cost_mapping_removed", &format!("{product}/{meter}"), &a.by).await?; |
| 1220 | return Ok(Outcome::Ok(CostMapping { |
| 1221 | product, |
| 1222 | meter, |
| 1223 | bucket: String::new(), |
| 1224 | price_meter: None, |
| 1225 | own_meter: None, |
| 1226 | scale_to_own: false, |
| 1227 | drift_percent: 0.0, |
| 1228 | note: String::new(), |
| 1229 | updated_at: now, |
| 1230 | updated_by: a.by, |
| 1231 | })); |
| 1232 | } |
| 1233 | let bucket = costs::slug(&a.bucket); |
| 1234 | if bucket.is_empty() { |
| 1235 | return Ok(Outcome::fail(FailureCode::Invalid, "Say which of g1t's products it is a cost of.")); |
| 1236 | } |
| 1237 | let clean = |v: Option<String>| v.map(|v| v.trim().to_owned()).filter(|v| !v.is_empty()); |
| 1238 | let (price_meter, own_meter) = (clean(a.price_meter), clean(a.own_meter)); |
| 1239 | let drift = a.drift_percent.filter(|d| d.is_finite() && *d > 0.0).unwrap_or(10.0); |
| 1240 | self.db |
| 1241 | .prepare( |
| 1242 | "INSERT INTO cost_map (product, meter, bucket, price_meter, own_meter, scale_to_own, drift_percent, note, updated_at, updated_by) |
| 1243 | VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10) |
| 1244 | ON CONFLICT (product, meter) DO UPDATE SET bucket = ?3, price_meter = ?4, own_meter = ?5, scale_to_own = ?6, |
| 1245 | drift_percent = ?7, note = ?8, updated_at = ?9, updated_by = ?10", |
| 1246 | ) |
| 1247 | .bind(&[ |
| 1248 | product.as_str().into(), |
| 1249 | meter.as_str().into(), |
| 1250 | bucket.as_str().into(), |
| 1251 | crate::optional(price_meter.as_deref()), |
| 1252 | crate::optional(own_meter.as_deref()), |
| 1253 | i32::from(a.scale_to_own).into(), |
| 1254 | drift.into(), |
| 1255 | a.note.trim().into(), |
| 1256 | now.as_str().into(), |
| 1257 | a.by.as_str().into(), |
| 1258 | ])? |
| 1259 | .run() |
| 1260 | .await?; |
| 1261 | self.audit("costs", "cost_mapping", &format!("{product}/{meter} → {bucket}"), &a.by).await?; |
| 1262 | Ok(Outcome::Ok(CostMapping { |
| 1263 | product, |
| 1264 | meter, |
| 1265 | bucket, |
| 1266 | price_meter, |
| 1267 | own_meter, |
| 1268 | scale_to_own: a.scale_to_own, |
| 1269 | drift_percent: drift, |
| 1270 | note: a.note.trim().to_owned(), |
| 1271 | updated_at: now, |
| 1272 | updated_by: a.by, |
| 1273 | })) |
| 1274 | } |
| 1275 | |
| 1276 | /// `admin_costs`: the Costs & margin page. |
| 1277 | pub(crate) async fn admin_costs(&self, a: AdminCostsArgs, configured: bool) -> Result<CostsReport> { |
| 1278 | let until = rfc3339(now_ms())[..10].to_owned(); |
| 1279 | let span = u64::from(a.days.unwrap_or(30).clamp(7, 90)); |
| 1280 | let since = day_before(&until, span - 1); |
| 1281 | let days = self.margin_days(&since, &until).await?; |
| 1282 | let rules = self.rules().await?; |
| 1283 | |
| 1284 | let mut products: BTreeMap<String, ProductMargin> = BTreeMap::new(); |
| 1285 | let mut overall = OverallMargin::default(); |
| 1286 | for d in &days { |
| 1287 | let p = products.entry(d.bucket.clone()).or_insert_with(|| ProductMargin { |
| 1288 | bucket: d.bucket.clone(), |
| 1289 | title: costs::bucket_title(&d.bucket), |
| 1290 | cost_source: if NOT_CLOUDFLARE.contains(&d.bucket.as_str()) { "ledger" } else { "cloudflare" }.into(), |
| 1291 | overhead: OVERHEAD.contains(&d.bucket.as_str()), |
| 1292 | ..ProductMargin::default() |
| 1293 | }); |
| 1294 | p.cf_cost_micros += d.cf_cost_micros; |
| 1295 | p.own_cost_micros += d.own_cost_micros; |
| 1296 | p.value_micros += d.value_micros; |
| 1297 | p.cost_micros += d.cost(); |
| 1298 | overall.cost_micros += d.cost(); |
| 1299 | if d.bucket == "platform" { |
| 1300 | overall.plans_micros += d.cash_micros; |
| 1301 | } else { |
| 1302 | overall.usage_micros += d.cash_micros; |
| 1303 | } |
| 1304 | } |
| 1305 | for p in products.values_mut() { |
| 1306 | p.margin_micros = p.value_micros - p.cost_micros; |
| 1307 | p.margin_percent = margin_percent(p.value_micros, p.cost_micros); |
| 1308 | } |
| 1309 | let revenue = overall.usage_micros + overall.plans_micros; |
| 1310 | overall.margin_micros = revenue - overall.cost_micros; |
| 1311 | overall.margin_percent = margin_percent(revenue, overall.cost_micros); |
| 1312 | let mut products: Vec<ProductMargin> = products.into_values().collect(); |
| 1313 | products.sort_by_key(|p| std::cmp::Reverse(p.cost_micros.max(p.value_micros))); |
| 1314 | |
| 1315 | #[derive(Deserialize)] |
| 1316 | struct DriftRow { |
| 1317 | bucket: String, |
| 1318 | kind: String, |
| 1319 | ours: f64, |
| 1320 | cloudflare: f64, |
| 1321 | delta_percent: Option<f64>, |
| 1322 | detail: String, |
| 1323 | found_at: String, |
| 1324 | } |
| 1325 | let drift = self |
| 1326 | .db |
| 1327 | .prepare("SELECT * FROM cost_drift ORDER BY kind, bucket") |
| 1328 | .all() |
| 1329 | .await? |
| 1330 | .results::<DriftRow>()? |
| 1331 | .into_iter() |
| 1332 | .map(|r| CostDrift { |
| 1333 | title: costs::bucket_title(&r.bucket), |
| 1334 | bucket: r.bucket, |
| 1335 | kind: r.kind, |
| 1336 | ours: r.ours, |
| 1337 | cloudflare: r.cloudflare, |
| 1338 | delta_percent: r.delta_percent, |
| 1339 | detail: r.detail, |
| 1340 | found_at: r.found_at, |
| 1341 | }) |
| 1342 | .collect(); |
| 1343 | |
| 1344 | #[derive(Deserialize)] |
| 1345 | struct Top { |
| 1346 | workspace: String, |
| 1347 | cost: Option<i64>, |
| 1348 | revenue: Option<i64>, |
| 1349 | internal: i64, |
| 1350 | } |
| 1351 | let top_workspaces = self |
| 1352 | .db |
| 1353 | .prepare(format!( |
| 1354 | "SELECT workspace, SUM(cost_micros) AS cost, SUM(revenue_micros) AS revenue, |
| 1355 | CASE WHEN workspace IN ({}) THEN 1 ELSE 0 END AS internal |
| 1356 | FROM workspace_costs WHERE day >= ?1 AND day <= ?2 GROUP BY workspace ORDER BY cost DESC LIMIT 15", |
| 1357 | crate::sales::INTERNAL_SQL |
| 1358 | )) |
| 1359 | .bind(&[since.as_str().into(), until.as_str().into()])? |
| 1360 | .all() |
| 1361 | .await? |
| 1362 | .results::<Top>()? |
| 1363 | .into_iter() |
| 1364 | .map(|t| WorkspaceCost { workspace: t.workspace, cost_micros: t.cost.unwrap_or(0), revenue_micros: t.revenue.unwrap_or(0), internal: t.internal == 1 }) |
| 1365 | .collect(); |
| 1366 | |
| 1367 | #[derive(Deserialize)] |
| 1368 | struct Summary { |
| 1369 | source: String, |
| 1370 | product: String, |
| 1371 | meter: String, |
| 1372 | raw_name: String, |
| 1373 | unit: String, |
| 1374 | quantity: f64, |
| 1375 | cost_usd: f64, |
| 1376 | } |
| 1377 | let lines = self |
| 1378 | .db |
| 1379 | .prepare( |
| 1380 | "SELECT source, product, meter, MAX(raw_name) AS raw_name, MAX(unit) AS unit, SUM(quantity) AS quantity, SUM(cost_usd) AS cost_usd |
| 1381 | FROM cost_lines WHERE day >= ?1 AND day <= ?2 GROUP BY source, product, meter ORDER BY cost_usd DESC, product, meter LIMIT 200", |
| 1382 | ) |
| 1383 | .bind(&[since.as_str().into(), until.as_str().into()])? |
| 1384 | .all() |
| 1385 | .await? |
| 1386 | .results::<Summary>()? |
| 1387 | .into_iter() |
| 1388 | .map(|l| CostLineSummary { |
| 1389 | bucket: costs::classify(&rules, &l.product, &l.meter).map(|r| r.bucket.clone()), |
| 1390 | product: l.product, |
| 1391 | meter: l.meter, |
| 1392 | raw_name: l.raw_name, |
| 1393 | unit: l.unit, |
| 1394 | source: l.source, |
| 1395 | quantity: l.quantity, |
| 1396 | cost_micros: micros(l.cost_usd), |
| 1397 | }) |
| 1398 | .collect(); |
| 1399 | |
| 1400 | #[derive(Deserialize)] |
| 1401 | struct MapRow { |
| 1402 | product: String, |
| 1403 | meter: String, |
| 1404 | bucket: String, |
| 1405 | price_meter: Option<String>, |
| 1406 | own_meter: Option<String>, |
| 1407 | scale_to_own: i64, |
| 1408 | drift_percent: f64, |
| 1409 | note: String, |
| 1410 | updated_at: String, |
| 1411 | updated_by: String, |
| 1412 | } |
| 1413 | let mappings = self |
| 1414 | .db |
| 1415 | .prepare("SELECT * FROM cost_map ORDER BY product, meter") |
| 1416 | .all() |
| 1417 | .await? |
| 1418 | .results::<MapRow>()? |
| 1419 | .into_iter() |
| 1420 | .map(|m| CostMapping { |
| 1421 | product: m.product, |
| 1422 | meter: m.meter, |
| 1423 | bucket: m.bucket, |
| 1424 | price_meter: m.price_meter, |
| 1425 | own_meter: m.own_meter, |
| 1426 | scale_to_own: m.scale_to_own == 1, |
| 1427 | drift_percent: m.drift_percent, |
| 1428 | note: m.note, |
| 1429 | updated_at: m.updated_at, |
| 1430 | updated_by: m.updated_by, |
| 1431 | }) |
| 1432 | .collect(); |
| 1433 | |
| 1434 | #[derive(Deserialize)] |
| 1435 | struct Fetched { |
| 1436 | at: Option<String>, |
| 1437 | } |
| 1438 | let fetched_at = self.db.prepare("SELECT MAX(fetched_at) AS at FROM cost_lines").first::<Fetched>(None).await?.and_then(|f| f.at); |
| 1439 | |
| 1440 | Ok(CostsReport { |
| 1441 | configured, |
| 1442 | fetched_at, |
| 1443 | days: days |
| 1444 | .iter() |
| 1445 | .map(|d| CostDay { |
| 1446 | day: d.day.clone(), |
| 1447 | bucket: d.bucket.clone(), |
| 1448 | cf_cost_micros: d.cf_cost_micros, |
| 1449 | own_cost_micros: d.own_cost_micros, |
| 1450 | value_micros: d.value_micros, |
| 1451 | cash_micros: d.cash_micros, |
| 1452 | }) |
| 1453 | .collect(), |
| 1454 | since, |
| 1455 | until, |
| 1456 | products, |
| 1457 | overall, |
| 1458 | drift, |
| 1459 | alerts: self.admin_cost_alerts(AdminCostAlertsArgs {}).await?, |
| 1460 | proposals: self.proposals().await?, |
| 1461 | versions: self.versions().await?, |
| 1462 | top_workspaces, |
| 1463 | lines, |
| 1464 | mappings, |
| 1465 | settings: self.cost_settings().await?, |
| 1466 | caps: self.spend_caps().await?, |
| 1467 | }) |
| 1468 | } |
| 1469 | } |
| 1470 | |
| 1471 | #[cfg(test)] |
| 1472 | mod tests { |
| 1473 | use super::*; |
| 1474 | |
| 1475 | #[test] |
| 1476 | fn the_overall_alert_says_dollars_while_little_comes_in() { |
| 1477 | let small = overall_detail(90_000, 7_500_000, 3, 10.0, -8239.7); |
| 1478 | assert!(small.contains("took in $0.09 against $7.50"), "{small}"); |
| 1479 | assert!(!small.contains('%'), "{small}"); |
| 1480 | let real = overall_detail(30_000_000, 40_000_000, 3, 10.0, -33.3); |
| 1481 | assert!(real.contains("as low as -33.3%"), "{real}"); |
| 1482 | } |
| 1483 | |
| 1484 | fn rule(product: &str, meter: &str, bucket: &str, own: Option<&str>) -> Rule { |
| 1485 | Rule { product: product.into(), meter: meter.into(), bucket: bucket.into(), price_meter: None, own_meter: own.map(Into::into), drift_percent: 10.0 } |
| 1486 | } |
| 1487 | |
| 1488 | fn rules() -> Vec<Rule> { |
| 1489 | vec![ |
| 1490 | rule("containers", "*", "sandboxes", None), |
| 1491 | rule("workers", "*", "platform", None), |
| 1492 | rule("artifacts", "*", "git", Some("git_operations")), |
| 1493 | rule("artifacts", "events_", "git", Some("git_operations")), |
| 1494 | ] |
| 1495 | } |
| 1496 | |
| 1497 | fn revenue_map() -> BTreeMap<String, String> { |
| 1498 | [("sandbox", "sandboxes"), ("git", "git"), ("plan", "platform")].iter().map(|(k, v)| (k.to_string(), v.to_string())).collect() |
| 1499 | } |
| 1500 | |
| 1501 | fn line(day: &str, source: &str, product: &str, meter: &str, quantity: f64, cost: f64) -> LineRow { |
| 1502 | LineRow { day: day.into(), source: source.into(), product: product.into(), meter: meter.into(), quantity, cost_usd: cost } |
| 1503 | } |
| 1504 | |
| 1505 | fn usage(day: &str, workspace: &str, key: &str, value: i64, cash: i64, cost: i64) -> UsageRow { |
| 1506 | UsageRow { day: day.into(), workspace: workspace.into(), key: key.into(), value, cash, cost } |
| 1507 | } |
| 1508 | |
| 1509 | #[test] |
| 1510 | fn a_day_puts_the_bill_g1ts_counts_and_charges_side_by_side() { |
| 1511 | let lines = vec![ |
| 1512 | line("2026-10-15", SOURCE_BILLABLE, "containers", "container_memory", 1000.0, 2.00), |
| 1513 | line("2026-10-15", SOURCE_BILLABLE, "artifacts", "artifacts_operations", 30_000.0, 3.00), |
| 1514 | // Artifacts' own events: not used while the bill has a count. |
| 1515 | line("2026-10-15", SOURCE_ARTIFACTS, "artifacts", "events_pull", 29_000.0, 0.0), |
| 1516 | line("2026-10-15", SOURCE_BILLABLE, "workers", "workers_cpu_ms", 1.0, 0.50), |
| 1517 | line("2026-10-15", SOURCE_BILLABLE, "browser_rendering", "browser_hours", 2.0, 0.25), |
| 1518 | ]; |
| 1519 | let own = vec![ |
| 1520 | OwnRow { day: "2026-10-15".into(), meter: "git_operations".into(), workspace: "acme".into(), quantity: 7_500.0 }, |
| 1521 | OwnRow { day: "2026-10-15".into(), meter: "git_operations".into(), workspace: "beta".into(), quantity: 2_500.0 }, |
| 1522 | ]; |
| 1523 | let usage = vec![ |
| 1524 | usage("2026-10-15", "acme", "sandbox", 2_400_000, 1_000_000, 2_000_000), |
| 1525 | usage("2026-10-15", "beta", "sandbox", 1_200_000, 1_200_000, 1_000_000), |
| 1526 | usage("2026-10-15", "acme", "git", 600_000, 600_000, 500_000), |
| 1527 | usage("2026-10-15", "acme", "implement", 120_000, 120_000, 100_000), |
| 1528 | usage("2026-10-15", "beta", "plan", 20_000_000, 20_000_000, 0), |
| 1529 | ]; |
| 1530 | let (days, workspaces) = fold(&rules(), &revenue_map(), &lines, &own, &usage); |
| 1531 | let get = |bucket: &str| days.iter().find(|d| d.bucket == bucket).unwrap(); |
| 1532 | let sandboxes = get("sandboxes"); |
| 1533 | assert_eq!((sandboxes.cf_cost_micros, sandboxes.own_cost_micros, sandboxes.value_micros, sandboxes.cash_micros), (2_000_000, 3_000_000, 3_600_000, 2_200_000)); |
| 1534 | let git = get("git"); |
| 1535 | assert_eq!(git.cf_cost_micros, 3_000_000); |
| 1536 | assert_eq!((git.cf_quantity, git.own_quantity), (30_000.0, 10_000.0)); |
| 1537 | assert_eq!(get("platform").value_micros, 20_000_000); |
| 1538 | // Not mapped: a leak until someone maps it. |
| 1539 | assert_eq!(get(UNMAPPED).cf_cost_micros, 250_000); |
| 1540 | // Models: no Cloudflare line, their cost is g1t's own. |
| 1541 | assert_eq!(get("models").cost(), 100_000); |
| 1542 | // Git's cost shared by g1t's own counts (Cloudflare gave none per |
| 1543 | // workspace here): three quarters to acme. |
| 1544 | let share = |ws: &str, bucket: &str| workspaces.iter().find(|w| w.workspace == ws && w.bucket == bucket).map(|w| (w.cost, w.revenue)); |
| 1545 | assert_eq!(share("acme", "git"), Some((2_250_000, 600_000))); |
| 1546 | assert_eq!(share("beta", "git"), Some((750_000, 0))); |
| 1547 | // Every bucket's cost is shared out exactly. |
| 1548 | for d in &days { |
| 1549 | let shared: i64 = workspaces.iter().filter(|w| w.bucket == d.bucket).map(|w| w.cost).sum(); |
| 1550 | assert_eq!(shared, d.cost(), "{}", d.bucket); |
| 1551 | } |
| 1552 | } |
| 1553 | |
| 1554 | #[test] |
| 1555 | fn artifacts_events_count_when_the_bill_does_not() { |
| 1556 | let lines = vec![ |
| 1557 | line("2026-10-05", SOURCE_ARTIFACTS, "artifacts", "events_pull", 120.0, 0.0), |
| 1558 | line("2026-10-05", SOURCE_ARTIFACTS, "artifacts", "events_push", 30.0, 0.0), |
| 1559 | line("2026-10-05", SOURCE_ARTIFACTS, "artifacts", "events_ratelimited", 9.0, 0.0), |
| 1560 | ]; |
| 1561 | let (days, _) = fold(&rules(), &revenue_map(), &lines, &[], &[]); |
| 1562 | assert_eq!(days[0].cf_quantity, 150.0); |
| 1563 | assert_eq!(days[0].cf_cost_micros, 0); |
| 1564 | } |
| 1565 | |
| 1566 | #[test] |
| 1567 | fn month_end_meters_are_told_by_the_day_from_snapshots() { |
| 1568 | let snap = |day: &str, cost: i64, charge: i64| (day.to_string(), "acme".to_string(), "git".to_string(), cost, charge); |
| 1569 | let rows = pending_deltas(&[snap("2026-10-30", 100, 120), snap("2026-10-31", 250, 300), snap("2026-11-01", 40, 48), snap("2026-11-02", 40, 48)]); |
| 1570 | assert_eq!( |
| 1571 | rows.iter().map(|r| (r.day.as_str(), r.cost, r.value)).collect::<Vec<_>>(), |
| 1572 | vec![("2026-10-30", 100, 120), ("2026-10-31", 150, 180), ("2026-11-01", 40, 48)] |
| 1573 | ); |
| 1574 | } |
| 1575 | |
| 1576 | #[test] |
| 1577 | fn a_plan_payment_is_spread_over_the_month_it_pays_for() { |
| 1578 | let days = spread("2026-10-01T00:00:00.000Z", 20_000_000, 30); |
| 1579 | assert_eq!(days.len(), 30); |
| 1580 | assert_eq!(days[0], ("2026-10-01".to_string(), 666_667)); |
| 1581 | assert_eq!(days[29], ("2026-10-30".to_string(), 666_666)); |
| 1582 | assert_eq!(days.iter().map(|d| d.1).sum::<i64>(), 20_000_000); |
| 1583 | assert!(spread("2026-10-01", 0, 30).is_empty()); |
| 1584 | assert_eq!(dollars(17_024_000), "$17.02"); |
| 1585 | assert_eq!(dollars(-27_668_620), "-$27.67"); |
| 1586 | assert_eq!(dollars(63_000), "$0.063"); |
| 1587 | } |
| 1588 | |
| 1589 | #[test] |
| 1590 | fn margins_and_deltas() { |
| 1591 | assert_eq!(margin_percent(1_200_000, 1_000_000).map(|m| (m * 100.0).round() / 100.0), Some(16.67)); |
| 1592 | assert_eq!(margin_percent(0, 5), None); |
| 1593 | assert_eq!(delta_percent(110.0, 100.0), Some(10.0)); |
| 1594 | assert_eq!(delta_percent(1.0, 0.0), None); |
| 1595 | } |
| 1596 | |
| 1597 | fn day(bucket: &str, cf: i64, own: i64, value: i64, cfq: f64, ownq: f64) -> ProductDay { |
| 1598 | ProductDay { day: "2026-10-15".into(), bucket: bucket.into(), cf_cost_micros: cf, own_cost_micros: own, value_micros: value, cash_micros: value, cf_quantity: cfq, own_quantity: ownq } |
| 1599 | } |
| 1600 | |
| 1601 | #[test] |
| 1602 | fn counts_more_than_the_threshold_apart_are_drift() { |
| 1603 | // Cloudflare counted 30,000 operations where g1t counted 10,000: |
| 1604 | // binding reads, perhaps. -66.7%. |
| 1605 | let drift = drifts("git", &[day("git", 3_000_000, 1_500_000, 1_800_000, 30_000.0, 10_000.0)], 10.0, true, 100_000); |
| 1606 | assert_eq!(drift.iter().map(|d| d.kind).collect::<Vec<_>>(), vec![DriftKind::Count, DriftKind::Cost]); |
| 1607 | assert!((drift[0].delta_percent.unwrap() + 66.666).abs() < 0.01); |
| 1608 | // 9% apart: within 10%. |
| 1609 | assert!(drifts("git", &[day("git", 1_000_000, 1_000_000, 1_200_000, 10_000.0, 10_900.0)], 10.0, true, 100_000).is_empty()); |
| 1610 | // Uncounted products have no count drift. |
| 1611 | assert!(drifts("sandboxes", &[day("sandboxes", 1_000_000, 1_050_000, 1_200_000, 5.0, 0.0)], 10.0, false, 100_000).is_empty()); |
| 1612 | } |
| 1613 | |
| 1614 | #[test] |
| 1615 | fn cost_with_no_revenue_is_a_leak_but_not_for_running_g1t() { |
| 1616 | let leak = drifts("actions_cache", &[day("actions_cache", 400_000, 0, 0, 0.0, 0.0)], 10.0, false, 100_000); |
| 1617 | assert_eq!(leak.len(), 1); |
| 1618 | assert_eq!(leak[0].kind, DriftKind::Leak); |
| 1619 | assert!(drifts("platform", &[day("platform", 5_000_000, 0, 0, 0.0, 0.0)], 10.0, false, 100_000).is_empty()); |
| 1620 | // Pennies say nothing. |
| 1621 | assert!(drifts("actions_cache", &[day("actions_cache", 50_000, 0, 0, 0.0, 0.0)], 10.0, false, 100_000).is_empty()); |
| 1622 | assert!(drifts(UNMAPPED, &[day(UNMAPPED, 250_000, 0, 0, 0.0, 0.0)], 10.0, false, 100_000)[0].kind == DriftKind::Leak); |
| 1623 | } |
| 1624 | |
| 1625 | #[test] |
| 1626 | fn a_margin_alert_needs_n_days_in_a_row_under_the_floor() { |
| 1627 | let s = |d: &str, revenue: i64, cost: i64| (d.to_string(), revenue, cost); |
| 1628 | // 5%, 0%, -20%: three days under 10%. |
| 1629 | let series = vec![s("10-13", 1_200_000, 1_000_000), s("10-14", 1_050_000, 1_000_000), s("10-15", 1_000_000, 1_000_000), s("10-16", 1_000_000, 1_200_000)]; |
| 1630 | let (from, worst) = breach(&series, 10.0, 3, 100_000).unwrap(); |
| 1631 | assert_eq!(from, "10-14"); |
| 1632 | assert!((worst + 20.0).abs() < 1e-9); |
| 1633 | // A good day in the window clears it. |
| 1634 | let mended = vec![s("10-14", 1_050_000, 1_000_000), s("10-15", 1_300_000, 1_000_000), s("10-16", 1_000_000, 1_200_000)]; |
| 1635 | assert!(breach(&mended, 10.0, 3, 100_000).is_none()); |
| 1636 | // Cost with no revenue at all is the worst margin there is. |
| 1637 | assert_eq!(breach(&[s("10-16", 0, 500_000)], 10.0, 1, 100_000).unwrap().1, -100.0); |
| 1638 | // Too little cost to judge. |
| 1639 | assert!(breach(&[s("10-16", 0, 5_000)], 10.0, 1, 100_000).is_none()); |
| 1640 | assert!(breach(&series, 10.0, 9, 100_000).is_none()); |
| 1641 | } |
| 1642 | |
| 1643 | #[test] |
| 1644 | fn shared_costs_add_up_to_the_bill() { |
| 1645 | let w = |k: &str, v: f64| (k.to_string(), v); |
| 1646 | assert_eq!(attribute(100, &[w("a", 1.0), w("b", 1.0), w("c", 1.0)]), vec![("a".into(), 34), ("b".into(), 33), ("c".into(), 33)]); |
| 1647 | assert_eq!(attribute(10, &[w("a", 3.0), w("b", 1.0), w("a", 0.0)]), vec![("a".into(), 8), ("b".into(), 2)]); |
| 1648 | assert!(attribute(10, &[w("a", 0.0)]).is_empty()); |
| 1649 | assert!(attribute(0, &[w("a", 1.0)]).is_empty()); |
| 1650 | } |
| 1651 | |
| 1652 | #[test] |
| 1653 | fn a_workspace_that_costs_more_than_it_pays_is_flagged() { |
| 1654 | let rows = vec![("acme".to_string(), 5_000_000, 1_000_000), ("beta".to_string(), 900_000, 0), ("gamma".to_string(), 2_000_000, 3_000_000)]; |
| 1655 | let found = anomalies(&rows, 1.0, 1_000_000); |
| 1656 | assert_eq!(found, vec![("acme".to_string(), 5_000_000, 1_000_000)]); |
| 1657 | // At twice its revenue as the threshold, $5 against $3 is fine. |
| 1658 | assert!(anomalies(&[("acme".to_string(), 5_000_000, 3_000_000)], 2.0, 1_000_000).is_empty()); |
| 1659 | } |
| 1660 | |
| 1661 | #[test] |
| 1662 | fn a_git_operation_costs_what_cloudflare_counts_for_it() { |
| 1663 | // $0.15 per 1,000 of Cloudflare's operations, on the charged days. |
| 1664 | let rate = billed_rate(&[(10_000.0, 0.0), (20_000.0, 3.0), (30_000.0, 4.5), (5_000.0, 0.75)]).unwrap(); |
| 1665 | assert!((rate - 0.000_15).abs() < 1e-12); |
| 1666 | // Cloudflare counted 3 for every 1 g1t did: binding reads count. |
| 1667 | let per_op = derived_unit_cost(rate, 300_000.0, 100_000.0).unwrap(); |
| 1668 | let per_thousand_micros = per_op * unit_size("1,000 operations") * 1e6; |
| 1669 | assert!((per_thousand_micros - 450_000.0).abs() < 1e-6, "{per_thousand_micros}"); |
| 1670 | // Too few of g1t's units to say. |
| 1671 | assert!(derived_unit_cost(rate, 3_000.0, 500.0).is_none()); |
| 1672 | assert!(billed_rate(&[(10_000.0, 0.0)]).is_none()); |
| 1673 | assert_eq!(unit_size("million requests"), 1e6); |
| 1674 | assert_eq!(unit_size("second"), 1.0); |
| 1675 | } |
| 1676 | } |