Bound the total number of Solr queries the API has in flight across concurrent requests

オープン
#311 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

まだ誰も着手していません。

評価

難易度
5/5
見積もり時間
1週間以上
初心者へのやさしさ
35/100
issue の種類
機能追加
明瞭さ
説明が足りない
活発さ
活発
技術スタック
python

調査の方向性

Start with api/server.py and Config.solr_max_concurrent_lookups, then review the shared httpx client and connection limits introduced by #310. Use the Solr capacity benchmark in TranslatorSRI/babel-validation#107 and the timing work in #309 to guide decisions about the process-wide limit, fairness, and overload behavior. Done means Solr concurrency is bounded across endpoints and concurrent requests, with queueing or rejection behavior and its timeout outcomes covered.

索引モデルが issue の本文から書いたものです。

説明

Nothing bounds how many Solr queries the NameRes API process has in flight in total.

  • SOLR_MAX_CONCURRENT_LOOKUPS (default 100) is a semaphore per /bulk-lookup request (Config.solr_max_concurrent_lookups in api/server.py). Its comment notes that the service-wide total is that limit multiplied by the number of bulk requests being served at once, and that the default of 100 "should be revisited once we know the real request rate", pointing at TranslatorSRI/babel-validation#107.
  • /lookup, /synonyms and the rest have no limit at all.
  • #310 replaces the per-query httpx clients with one shared client per event loop. It deliberately leaves that client's pool uncapped (httpx.Limits(max_connections=None, ...)) to keep today's behaviour. A cap below what /bulk-lookup can have in flight would make queries wait for a pool slot inside SOLR_TIMEOUT_SECONDS, so a busy pool would show up as Solr timeouts.

So a few concurrent large bulk requests, or a burst of autocomplete traffic, can put hundreds of queries on Solr at once. #306 (six concurrent lookups for "vitamin a", each blocked on Solr for 4-5 minutes) may be this or something else, but nothing in the API stops a pile-up like it.

Questions to settle:

  • Where the bound lives. A process-wide semaphore around every Solr request, a capped httpx pool, or both. With a capped pool, the wait for a connection needs its own timeout (httpx's pool timeout), separate from Solr's, so that recent_queries can tell "NameRes queued it" apart from "Solr was slow". That probably means another outcome alongside timeouts.
  • Fairness. Whether one /bulk-lookup can take every slot, starving /lookup (interactive autocomplete), or whether bulk gets a share.
  • The number. It should come from Solr's capacity under the measured request mix (babel-validation#107's benchmark), not a guess. It also needs dividing by the number of API replicas, since each replica has its own limit.
  • What callers see when it's full: queueing, or a fast 503 with Retry-After.

Related: #309 (timing /bulk-lookup, which would show how long bulk strings wait for a slot).

主要言語
Python
スター
5
フォーク
4
PR マージ指標
30日以内にマージされた PR はありません

環境構築

  • Dockerfile または Docker Compose ファイルあり
  • プルリクエストのテンプレートなし
  • コントリビューションガイドなし

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

NCATSTranslator/NameResolution のほかの issue

NCATSTranslator/NameResolution の issue をすべて見る

似ている issue

Python の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。