Embedded Replica update fails

[khuezy]

When using Embedded Replicas, calling the UPDATE api results in Hrana 404 "stream not found"

@libsql/client: 0.15.7

Edit: turso dies after a few minutes... even GET requests are 404ing.

Edit2: 0.15.8 did not fix

[penberg]

My current hypothesis is that we keep a SQL over HTTP connection open for too long in the embedded replica internals and it expires.

[khuezy]

Can you cut a prerelease with that fix for me to test?
And why didn't this happen on Fly.io?

[khuezy]

I'm also getting wal_insert_begin failed when writing. Is this happening when the syncInterval is running and locking the database when I try to write?

[khuezy]

@penberg @avinassh
https://github.com/khuezy/next-serwist-bug
Reproduction in the readme.

[khuezy]

Per a discussion on discord, when setting syncInterval to a lower value, eg 3 seconds, it would throw the wal_insert_begin failed error frequently.
It appears that the client is stepping on its own toes when locking, reading/writing on the local file...

[e-cal]

You can work around the issue by modifying the pool parameters, e.g. based on #1203 (comment)

from sqlalchemy.pool import QueuePool

engine = create_engine(
    db_url,
    poolclass=QueuePool,
    pool_size=5,
    max_overflow=10,
    pool_timeout=30,
    pool_recycle=1800,
    connect_args={
        "check_same_thread": False,
        "auth_token": AUTH_TOKEN,
    },
    echo=False,
)

[khuezy]

I don't think those options are exposed in the node client.