762a240890
- Log system revamp: - Logs to stdout, disabled while testing - Use header `X-Request-Id`, or create a new `uuid` when no present, to identyfy log entries - Be able to set log level from env variable `LOG_LEVEL`, useful while testing: `LOG_LEVEL=info npm test`; even more human-readable: `LOG_LEVEL=info npm t | ./node_modules/.bin/pino-pretty` - Be able to reduce the footprint in the final log file depending on the environment - Use one logger for every service: Queries, Batch Queries (Jobs), and Data Ingestion (CopyTo/CopyFrom) - Stop using headers such as: `X-SQL-API-Log`, `X-SQL-API-Profiler`, and `X-SQL-API-Errors` as a way to log info. - Be able to tag requests with labels as an easier way to provide business metrics - Metro: Add log-collector utility (`metro`), it will be moved to its own repository. Attaching it here fro development purposes. Try it with the following command `LOG_LEVEL=info npm t | node metro` - Metro: Creates `metrics-collector.js` a stream to update Prometheus' counters and histograms and exposes them via Express' app (`:9145/metrics`). Use the ones defined in `grok_exporter` Announcements: - Profiler is always set. No need to check its existence anymore - Unify profiler usage for every endpoint Bug fixes: - Avoid hung requests while fetching user identifier |
||
---|---|---|
.. | ||
leader | ||
maintenance | ||
models | ||
pubsub | ||
scheduler | ||
util | ||
batch.js | ||
index.js | ||
job-backend.js | ||
job-canceller.js | ||
job-queue.js | ||
job-runner.js | ||
job-service.js | ||
job-status.js | ||
query-runner.js | ||
README.md | ||
user-database-metadata-service.js |
Batch Queries
This document describes features from Batch Queries, it also details some internals that might be useful for maintainers and developers.
Redis data structures
Jobs definition
Redis Hash: batch:jobs:{UUID}
.
Redis DB: 5.
It stores the job definition, the user, and some metadata like the final status, the failure reason, and so.
Job queues
Redis List: batch:queue:{username}
.
Redis DB: 5.
It stores a pending list of jobs per user. It points to a job definition with the {UUID}
.
Job notifications
Redis Pub/Sub channel: batch:users
.
Redis DB: 0.
In order to notify new jobs, it uses a Pub/Sub channel were the username for the queued job is published.
Job types
Format for the currently supported query types, and what they are missing in terms of features.
Simple
{
"query": "update ..."
}
Does not support main fallback queries. Ideally it should support something like:
{
"query": "update ...",
"onsuccess": "select 'general success fallback'",
"onerror": "select 'general error fallback'"
}
Multiple
{
"query": [
"update ...",
"select ... into ..."
]
}
Does not support main fallback queries. Ideally it should support something like:
{
"query": [
"update ...",
"select ... into ..."
],
"onsuccess": "select 'general success fallback'",
"onerror": "select 'general error fallback'"
}
Fallback
{
"query": {
"query": [
{
"query": "select 1",
"onsuccess": "select 'success fallback query 1'",
"onerror": "select 'error fallback query 1'"
},
{
"query": "select 2",
"onerror": "select 'error fallback query 2'"
}
],
"onsuccess": "select 'general success fallback'",
"onerror": "select 'general error fallback'"
}
}
It's weird to have two nested query
attributes. Also, it's not possible to mix plain with fallback ones.
Ideally it should support something like:
{
"query": [
{
"query": "select 1",
"onsuccess": "select 'success fallback query 1'",
"onerror": "select 'error fallback query 1'"
},
"select 2"
],
"onsuccess": "select 'general success fallback'",
"onerror": "select 'general error fallback'"
}
}
Where you don't need a nested query
attribute, it's just an array as in Multiple job type, and you can mix objects and
plain queries.