Files
mtusi2026lab2/scripts/track2/run.sh
T
2026-10-23 21:13:27 +03:00

147 lines
9.1 KiB
Bash
Executable File
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
#!/bin/sh
# Трек 2: план запроса, стоимость индексов и backup/restore. Изолированный стенд: свои контейнеры, сеть и тома track2-*;
# стенд compose (messenger-lab2) не затрагивается. Фазы: up | indexes | backup | down. Сырые результаты — в evidence/track2/.
# sh scripts/track2/run.sh up # контейнер PostgreSQL, миграции приложения, 115 000 сообщений
# sh scripts/track2/run.sh indexes # сравнение вариантов индекса: планы, размер, pgbench записи/чтения
# sh scripts/track2/run.sh backup # pg_dump -> восстановление в новое хранилище -> сверка
# sh scripts/track2/run.sh down # удалить всё
# Требуются Docker, .NET SDK (dotnet tool restore уже выполнен), образ messenger-api:lab2 (docker compose build) для фазы backup.
set -eu
cd "$(dirname "$0")/../.."
export MSYS_NO_PATHCONV=1 # Git Bash на Windows не должен переписывать пути внутри docker exec
OUT=evidence/track2
NET=track2net
SRC=track2-src
RST=track2-restore
IMG=$(grep -o 'postgres:17@sha256:[0-9a-f]*' compose.yaml | head -1)
WORK=scripts/track2/.work # относительный путь: его понимают и dotnet.exe, и POSIX-оболочка
mkdir -p "$OUT" "$WORK"
. scripts/track2/backup.sh
src_psql() { docker exec -i "$SRC" psql -U postgres -d messenger -v ON_ERROR_STOP=1 "$@"; }
scalar() { src_psql -At -c "$1"; }
wait_ready() { # $1 — контейнер
for _ in $(seq 1 60); do
docker exec "$1" pg_isready -U postgres -d messenger >/dev/null 2>&1 && return 0
sleep 1
done
echo "контейнер $1 не готов" >&2; exit 1
}
phase_up() {
[ -f "$WORK/pw" ] || LC_ALL=C tr -dc 'A-Za-z0-9' < /dev/urandom | head -c 24 > "$WORK/pw"
docker network inspect "$NET" >/dev/null 2>&1 || docker network create "$NET" >/dev/null
docker volume create track2_src >/dev/null
docker run -d --name "$SRC" --network "$NET" -e POSTGRES_PASSWORD="$(cat "$WORK/pw")" -e POSTGRES_DB=messenger \
-v track2_src:/var/lib/postgresql/data "$IMG" >/dev/null
wait_ready "$SRC"; sleep 2; wait_ready "$SRC"
# Схема — те же миграции EF Core, что применяет приложение.
dotnet tool run dotnet-ef migrations script --project scr/Messenger.Api -o "$WORK/schema.sql" >/dev/null
src_psql -q < "$WORK/schema.sql"
echo "seed..."; src_psql -q < scripts/track2/seed.sql
src_psql -c "SELECT (SELECT count(*) FROM messages) AS messages, (SELECT count(*) FROM chats) AS chats, (SELECT max(c) FROM (SELECT count(*) c FROM messages GROUP BY chat_id) s) AS biggest_chat" \
| tee "$OUT/seed-summary.txt"
}
# Варианты индекса истории: имя | определение.
create_variant() {
case "$1" in
none) : ;;
hist) src_psql -q -c "CREATE INDEX ix_t2_hist ON messages (chat_id, created_at, id)" ;;
nosort) src_psql -q -c "CREATE INDEX ix_t2_nosort ON messages (chat_id, created_at)" ;;
covering) src_psql -q -c "CREATE INDEX ix_t2_covering ON messages (chat_id, created_at, id) INCLUDE (sender_id, text)" ;;
esac
}
bench_lines() { # tps и средняя задержка из вывода pgbench
grep -E "^(tps|latency average|number of transactions actually processed)" | tr '\n' ' '; echo
}
phase_indexes() {
big=$(scalar "SELECT big FROM bench.params"); medium=$(scalar "SELECT medium FROM bench.params")
small=$(scalar "SELECT small FROM bench.params"); sender=$(scalar "SELECT sender FROM bench.params")
ncur=$(scalar "SELECT count(*) FROM bench.cursors")
mid_t=$(scalar "SELECT t FROM bench.cursors WHERE n = (SELECT count(*)/2 FROM bench.cursors)")
mid_id=$(scalar "SELECT id FROM bench.cursors WHERE n = (SELECT count(*)/2 FROM bench.cursors)")
end_t=$(scalar "SELECT t FROM bench.cursors WHERE n = (SELECT max(n) FROM bench.cursors)")
end_id=$(scalar "SELECT id FROM bench.cursors WHERE n = (SELECT max(n) FROM bench.cursors)")
docker cp scripts/track2/write.pgbench "$SRC":/tmp/write.pgbench
docker cp scripts/track2/read.pgbench "$SRC":/tmp/read.pgbench
: > "$OUT/variants-summary.txt"
for v in none hist nosort covering; do
echo "=== variant: $v"
# Чистое состояние: удалить строки прошлых прогонов записи, пересобрать таблицу, оставить только индексы варианта.
src_psql -q -c "DELETE FROM messages WHERE text LIKE 'bench %'"
src_psql -q -c "VACUUM FULL messages"
src_psql -q -c "DROP INDEX IF EXISTS ix_messages_chat_history, ix_t2_hist, ix_t2_nosort, ix_t2_covering"
create_variant "$v"
src_psql -q -c "ANALYZE messages"
size=$(scalar "SELECT coalesce((SELECT pg_size_pretty(pg_relation_size(c.oid)) FROM pg_class c WHERE c.relname = 'ix_t2_$v'), '-')")
# Планы снимаются на втором (прогретом) запуске: первый только заполняет shared buffers.
for pass in 1 2; do
src_psql -q -v big="$big" -v small="$small" -v mid_t="$mid_t" -v mid_id="$mid_id" -v end_t="$end_t" -v end_id="$end_id" \
-f - < scripts/track2/plans.sql > "$OUT/plans-$v.txt"
done
# Индекс с INCLUDE (text) ограничен размером строки B-дерева (~2700 байт): проверяем сообщение максимальной длины 4000 символов.
limit_msg=$(src_psql -At -c "BEGIN; INSERT INTO messages (id, chat_id, sender_id, text, created_at) SELECT gen_random_uuid(), '$small', '$sender', (SELECT string_agg(chr(33 + (random()*93)::int), '') FROM generate_series(1,4000)), now(); ROLLBACK;" 2>&1 | grep -E "ERROR|INSERT" | head -1 || true)
echo "read:" > "$WORK/bench.tmp"
for i in 1 2 3; do
docker exec "$SRC" pgbench -U postgres -d messenger -n -c 4 -j 2 -T 10 -f /tmp/read.pgbench \
-D big="'$big'" -D ncur="$ncur" 2>&1 | bench_lines >> "$WORK/bench.tmp"
done
echo "write:" >> "$WORK/bench.tmp"
lsn0=$(scalar "SELECT pg_current_wal_lsn()")
for i in 1 2 3; do
docker exec "$SRC" pgbench -U postgres -d messenger -n -c 4 -j 2 -T 10 -f /tmp/write.pgbench \
-D big="'$big'" -D medium="'$medium'" -D sender="'$sender'" 2>&1 | bench_lines >> "$WORK/bench.tmp"
done
# Размер индекса после нагрузки записи — до bulk-вставки: её откат оставляет мёртвые записи в индексе.
idx_after=$(scalar "SELECT coalesce((SELECT pg_size_pretty(pg_relation_size(c.oid)) FROM pg_class c WHERE c.relname = 'ix_t2_$v'), '-')")
wal_bytes=$(scalar "SELECT pg_wal_lsn_diff(pg_current_wal_lsn(), '$lsn0')")
wal_rows=$(scalar "SELECT count(*) FROM messages WHERE text LIKE 'bench %'")
wal_per_row=$(awk "BEGIN { printf \"%.0f\", $wal_bytes / $wal_rows }")
# Bulk-вставка 100 000 строк в большой чат одним оператором (без fsync на каждую строку): чистая стоимость поддержки индекса.
bulk=$(src_psql -At -c "BEGIN; EXPLAIN (ANALYZE, WAL, COSTS OFF, TIMING OFF) INSERT INTO messages (id, chat_id, sender_id, text, created_at) SELECT gen_random_uuid(), '$big', '$sender', 'bulk ' || g, timestamptz '2027-01-01' + g * interval '1 second' FROM generate_series(1, 100000) g; ROLLBACK;" | grep -E "Execution Time|WAL:" | tr '
' ' ')
{
echo "### $v"
echo "index_size_before_writes: $size ; index_size_after_writes: $idx_after"
echo "insert_4000_chars: ${limit_msg:-ok}"
echo "wal_bytes_per_inserted_row (pgbench, 1 строка = 1 транзакция): $wal_per_row"
echo "bulk_insert_100000_rows: $bulk"
cat "$WORK/bench.tmp"
echo
} | tee -a "$OUT/variants-summary.txt"
cp "$WORK/bench.tmp" "$OUT/pgbench-$v.txt"
done
# Вернуть канонический индекс приложения (как в миграции), убрать строки нагрузки.
src_psql -q -c "DELETE FROM messages WHERE text LIKE 'bench %'"
src_psql -q -c "VACUUM FULL messages"
src_psql -q -c "DROP INDEX IF EXISTS ix_t2_hist, ix_t2_nosort, ix_t2_covering"
src_psql -q -c "CREATE INDEX ix_messages_chat_history ON messages (chat_id, created_at, id)"
src_psql -q -c "ANALYZE"
}
phase_down() {
docker rm -f track2-api-src track2-api-restore "$RST" "$SRC" >/dev/null 2>&1 || true
docker volume rm track2_src track2_restore >/dev/null 2>&1 || true
docker network rm "$NET" >/dev/null 2>&1 || true
rm -rf "$WORK"
}
case "${1:-}" in
up) phase_up ;;
indexes) phase_indexes ;;
backup) phase_backup ;;
down) phase_down ;;
*) echo "usage: $0 up|indexes|backup|down" >&2; exit 2 ;;
esac