File size: 12,043 Bytes
44084af | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 | """Text for triage items (R08, R10, R11, R12, R13, R18) and for the complaint about an alert that was right.
In E5 the triage items are answers in the on-call survey: first person, written by whoever was annoyed, not
tickets (R2). Everywhere else they are queue tickets."""
from __future__ import annotations
import datetime as dt
from . import evidence as ev
from . import variety
from .items import Ticket
from .people import crew
from .prose_svc import _day, _key, paged_team
from .xscen import ip, pods, series
def _survey(w, item) -> bool:
return w.family == "E5" and item.code in variety.SURVEY_CODES
def generic_runbook(alert: str, w=None) -> str:
opts = [f"""---
alerts: [{alert}]
---
# {alert}
1. Check the service dashboard (svc-overview) for the affected service.
2. Look at recent deploys (`#deploys`) before digging further.
3. If the database is involved, loop in database-reliability.
""", f"""---
alerts: [{alert}]
---
# {alert}
Start from svc-overview filtered to the service. Most of the time something was just deployed: check #deploys
and the rollout history. Database trouble goes to database-reliability, they have their own rota.
""", f"""---
alerts: [{alert}]
---
# {alert}
- service dashboard first (svc-overview, pick the service)
- anything in #deploys in the last hour? roll it back before investigating
- DB involved: page database-reliability
"""]
return variety.card(w, "any:R13:runbook", opts) if w is not None else opts[0]
def r08(item, w, rng) -> Ticket:
z, s = item.p["gone"], item.p["svc"]
title, body = variety.card(w, "R08", [
(f"{z} is gone", f"`{z}` was switched off on {item.p['date']}; its alerts and its route block are still in "
f"here. Please clean them out. `{s}` stays, obviously."),
(f"remove {z}", f"Decommissioned on {item.p['date']} ({item.p['team']} confirmed). Whatever still refers to "
f"it in rules and routing can go. Don't touch `{s}`."),
(f"{z} leftovers", f"{z} has had zero replicas since {item.p['date']} and the deployment is being deleted. "
f"Its rules and routing are still in this repo. {s} is the replacement and stays.")])
return Ticket(_key(w, rng), title, body, crew(w, rng, item.p["team"], 1)[0])
def severity(item, w, rng) -> Ticket:
a = item.p["alert"]
who = crew(w, rng, paged_team(w, item.p["svc"]), 1)[0]
if item.direction == "up":
other = rng.choice([s.name for s in w.services if s.kind == "api"] or ["the API"])
if _survey(w, item):
text = variety.card(w, "survey:R10", [
f"{a} should page. Twice now it was the first sign of trouble and both times it sat in Slack for twenty "
f"minutes before anyone looked. Customers were unable to complete requests while it was firing.",
f"I keep seeing {a} in Slack hours after the fact. It warned us before the backlog took {other} down "
f"last month and nobody was paged. Customers could not complete requests. Make it a page.",
f"Can we finally make {a} page? Customers cannot complete requests while it is firing. "
f"We said so in the ops review and it still only posts to Slack."])
return Ticket(_key(w, rng), f"make {a} page", text, who)
title, body = variety.card(w, "R10", [
(f"make {a} page", f"{a}: make it page. It caught the last two incidents before customers did, and both "
f"times it sat in Slack for twenty minutes before anyone looked."),
(f"{a} should wake someone up", f"Can {a} page please? It's the only thing that warned us before the "
f"backlog took {other} down last month."),
(f"{a} to page", f"We agreed in the ops review that {a} should wake someone up. Right now it only posts to "
f"Slack.")])
return Ticket(_key(w, rng), title, body, who)
if _survey(w, item):
text = variety.card(w, "survey:R11", [
f"{a} pages whoever's on call at 3 a.m. and there is nothing I can do about it until the upstream team is "
f"in. It should be a ticket for the owning team.",
f"Third night this month {a} paged me and the answer was \"wait for business hours\". That's a ticket, "
f"not a page.",
f"Nobody can act on {a} at night. I ack it and go back to sleep. Put it in the team's queue instead."])
return Ticket(_key(w, rng), f"{a} shouldn't page", text, who)
title, body = variety.card(w, "R11", [
(f"{a} shouldn't page", f"{a} pages whoever's on call at 3 a.m. and there is nothing they can do about it until "
f"the upstream team is in. It should go to the team's queue instead."),
(f"{a}: ticket, not page", f"Nobody can act on {a} at night. It should be a ticket for the owning team, not a "
f"page."),
(f"stop paging for {a}", f"Third night this month {a} paged and the answer was \"wait for business hours\". "
f"Make it a ticket.")])
return Ticket(_key(w, rng), title, body, who)
def r12(item, w, rng) -> Ticket:
a = item.p["alert"]
who = crew(w, rng, paged_team(w, item.p["svc"]), 1)[0]
if _survey(w, item):
text = variety.card(w, "survey:R12", [
f"The Slack message for {a} just says \"p99 latency above 500ms on\" and then nothing. Which pod? Which "
f"service? I have to open Prometheus to find out.",
f"Small one: {a} notifications end mid-sentence (\"... on\"). I'd like to know what they're about without "
f"clicking through.",
f"Got {a} in Slack at night: \"p99 latency above 500ms on\". On what? I spent five minutes finding out."])
return Ticket(_key(w, rng), f"{a} message is cut off", text, who)
title, body = variety.card(w, "R12", [
(f"{a} message is cut off", f"The Slack message for {a} just says \"p99 latency above 500ms on\" and then "
f"nothing. Which pod? Which service? I have to open Prometheus to find out."),
(f"{a} notifications end mid-sentence", f"{a} notifications end mid-sentence (\"... on\"). Please make them say "
f"what they're about."),
(f"{a}: summary doesn't say what's slow", f"Got {a} in Slack last night: \"p99 latency above 500ms on\". On "
f"what? Summary needs to name the thing that's slow.")])
return Ticket(_key(w, rng), title, body, who)
def r13(item, w, rng) -> Ticket:
a = item.p["alert"]
who = crew(w, rng, "sre", 1)[0]
if _survey(w, item):
text = variety.card(w, "survey:R13", [
f"The runbook link on {a} goes to the old wiki, which 404s since the migration. There's a runbook for it "
f"in the repo, I checked.",
f"Clicked the runbook on {a} mid-incident and got the old Confluence 404. Fun.",
f"Please go through the runbook links. {a} still points at the old wiki, which is gone."])
return Ticket(_key(w, rng), f"{a} runbook link 404s", text, who)
title, body = variety.card(w, "R13", [
(f"{a} runbook link 404s", f"The runbook link on {a} goes to the old wiki, which 404s since the migration. "
f"There is a runbook for it in the repo."),
(f"dead runbook link on {a}", f"{a}'s runbook link is dead (old wiki). Point it where the rest of them point."),
(f"{a}: runbook goes to the old wiki", f"Clicked the runbook on {a} mid-incident and got the old Confluence "
f"404. Please fix the link.")])
return Ticket(_key(w, rng), title, body, who)
def r18(item, w, rng) -> Ticket:
who = crew(w, rng, "sre", 1)[0]
if _survey(w, item):
text = variety.card(w, "survey:R18", [
f"Why do I get {item.NEW} and {item.OLD} for the same disk, every single time? {item.OLD} is the old one.",
f"{item.OLD} and {item.NEW} always fire together. One of them has to go, and {item.OLD} is the older one.",
f"Two alerts for one full disk ({item.OLD} and {item.NEW}). I'd keep {item.NEW}."])
return Ticket(_key(w, rng), "duplicate disk alerts", text, who)
title, body = variety.card(w, "R18", [
("duplicate disk alerts", f"We get {item.NEW} and {item.OLD} for the same thing, every time. {item.OLD} is the "
f"old one."),
(f"drop {item.OLD}", f"{item.OLD} and {item.NEW} always fire together. Drop {item.OLD}, it's the older of the two."),
(f"{item.OLD} duplicates {item.NEW}", f"Every disk alert arrives twice, once as {item.OLD} and once as "
f"{item.NEW}. {item.NEW} is the newer one; keep that and drop the other.")])
return Ticket(_key(w, rng), title, body, who)
def kept(k, w, rng) -> Ticket:
s = k.svc
date = _day(w, rng, 2, 16)
clock = ev.Clock(date, "02:40", w.company.utc_offset, w.company.tz_label)
n = 60
act = [int(v) for v in [60 + (m * 0.4 if m < 20 else 8 + (m - 20) * 2.2) for m in range(n + 1)]]
act = [min(v, 97) if m < 45 else 30 for m, v in enumerate(act)]
pod = pods(rng, s.name, 1)[0]
lab = {"service": s.name, "pod": pod, "instance": f"{ip(rng)}:8080", "namespace": w.namespace}
raws = [series("db_pool_connections_active", lab, act), series("db_pool_connections_max", lab, [100] * (n + 1))]
path = f"incidents/{w.cal.inc(date, '02:40', 30)}"
peak = next((m for m, v in enumerate(act) if v >= 96), 30)
files = {f"{path}/query-range/{s.name}-db-pool.json": ev.query_range(raws, clock),
f"{path}/app.log": app_log(rng, clock, s.name, pod, peak)}
text = variety.card(w, "kept", [
f"{k.alert} paged me at {clock.hm(peak - 5)} for nothing. Nobody complained, it resolved by itself. "
f"Can we make it less jumpy? Pool numbers and the app log from that night are in `{path}/`.",
f"Woken up by {k.alert} at {clock.hm(peak - 5)}. I don't think it was real, the service looked fine by the "
f"time I had a laptop open. Stuff from that night: `{path}/`.",
f"I got {k.alert} at {clock.hm(peak - 5)} and by the time my laptop was open {s.name} was fine. Was it ever "
f"real? Pool graph and logs: `{path}/`."])
return Ticket(_key(w, rng), f"{k.alert} woke me up", text, crew(w, rng, paged_team(w, s.name), 1)[0], files)
def app_log(rng, clock, svc: str, pod: str, peak: int) -> str: # noqa: ARG001
"""The app's log around the pool running dry (R8): bursts of failed requests with request ids, warnings about
slow checkouts before it, ordinary lines around it."""
lines = []
t = (peak - 6) * 60.0
while t < (peak + 6) * 60:
sec = rng.uniform(0.2, 9.0) if peak * 60 - 30 < t < peak * 60 + 120 else rng.uniform(4, 40)
t += sec
stamp = (clock.local(0) + dt.timedelta(seconds=t)).strftime("%Y-%m-%dT%H:%M:%S.%f")[:-3]
rid = "".join(rng.choice("0123456789abcdef") for _ in range(16))
m = t / 60
if peak - 1 <= m < peak + 3 and rng.random() < 0.7:
lines.append(f"{stamp} ERROR [{svc},{rid}] pool: FATAL: sorry, too many clients already")
elif peak - 4 <= m < peak + 3 and rng.random() < 0.5:
lines.append(f"{stamp} WARN [{svc},{rid}] pool: connection checkout took {rng.uniform(1.2, 4.8):.1f}s "
f"(active={rng.randint(88, 99)}, max=100)")
else:
lines.append(f"{stamp} INFO [{svc},{rid}] {rng.choice(['GET', 'GET', 'POST'])} /v1/{rng.choice(['items', 'orders', 'search', 'accounts'])} "
f"{rng.choice([200, 200, 200, 204, 404])} {rng.randint(8, 240)}ms")
return "\n".join(lines) + "\n"
|