+1 I'm also keen for this to be fixed. It has flow on effects in drizzle-orm, joins just don't work
+1 I'm also keen for this to be fixed. It has flow on effects in drizzle-orm, joins just don't work correctly for any queries run against D1.
MI can send you a temporary fix if you are stuck with it
ANah that's ok, I've worked around it by not using joins for now. But it's definitely something I have to actively work around, so keen for a fix from the Cloudflare team.
M
MPlease don't do this.
SIt looks like TS bindings don't align with actual types when calling 'all' from a prepared statement 
The 'results' field has a type of T[][], when actually it's just T[], so I'm forced to use ts-expect-error. Using version 4.20230628.0 of workers types, but it was the case at least since 4.20230321.0 (that's when I started using D1). Also, I'm using experimental backend for D1
Am I doing something wrong? If not, how do I report it properly (e.g. github issue) so it can be tracked?
The 'results' field has a type of T[][], when actually it's just T[], so I'm forced to use ts-expect-error. Using version 4.20230628.0 of workers types, but it was the case at least since 4.20230321.0 (that's when I started using D1). Also, I'm using experimental backend for D1
Am I doing something wrong? If not, how do I report it properly (e.g. github issue) so it can be tracked?
IIt seems like itās passing in Out[] as the result type, which is not what you are passingā¦ have you tried restarting Typescript or the editor?
UFrom reading the issue it seems like they are probably just going to make that throw an error instead of returning data at all and you should be using
AS to rename duplicated columns https://github.com/cloudflare/workerd/pull/696#issuecomment-1561891627
VDurable Objects run in a ton of regions, as for the D1 read replicas will they match these regions or run in fewer ones? DO give the best performance as of now as they run the closest and was just wondering if this will be the same in the future. Guess it's kinda overkill to run 8 read replicas for a db.
UD1 is a durable object so it will be the same regions but how many it will be in no clue. My guess is probably 3 maybe 4 
UDurable objectās performance really depend though because if you have 1 object used by many users it wonāt be close to all of them as they arenāt replicated each object only exists in one location
VI have one DO per user and this is why I was wondering about D1/DO speed
Vas with my current setup DO is really close to the user and so everything is snappy. D1 will be single-writer as stated by the team so the write speeds will be always faster for DO, but I guess reads will be at least as good or better for DOs too.
VIt's just that I'd love to write SQL instead of DOs, lot easier for other developers in the team too.
VAnd so which one does one choose, a few hundered ms faster writes, and a couple ten ms faster reads for the end users or easier developer experience
Veasier migrations etc etc, all the goodies that come with the abstraction of D1
MMost web applications are (at least) 80:20 read:write, for what itās worth.
MNothing stops you from creating multiple D1 databases either ā but if you want to query all of your (global) data you need to make a trade-off. We donāt think D1 and DO are really comparable: a SQL database vs a strongly consistent key:value store.
VYeah multiple databases is the same tradeoff as with Durable Objects, I need a global singleton instance to keep track of like list of all users in the system and since that is a single DO it can increase latency for some endpoint where accessing it is needed.
VAlso, even if 80:20 is the ratio of read:writes I have a hard time justifying a 500ms write response for globally distributed users. I know going single writer gives us a lot of consistency but honestly I really like Dynamos model, where one at least has the option to go multi writer while still having a single database/table
VAnyways, can't wait for D1 to get out of beta and to learn about it's capabilities and test it against other DBs 
KHey! Just want to let you know I'm continuing to try to debug what's going on with your database. It seems for some reason our metadata storage is dropping connections but only when we try to query your object specifically. The data is all there and fine, but the backend drops the connection whenever we try to grab the lock to start up the object, and our current observability isn't sufficient to tell me what exactly went wrong. Debugging is slow because we have to push a new runtime release to test each theory about what's happening here, since we can repro with any kind of test data, but it's going... (I did three releaess over the weekend for this.)
It's possible we could get the object working again more quickly by resetting some state but then we'd never know exactly what went wrong, so it could happen again someday. Since you say it's not critical for you, I'd like to keep trying to debug, but let me know if that changes and you want me to prioritize getting the object working without understanding what broke.
It's possible we could get the object working again more quickly by resetting some state but then we'd never know exactly what went wrong, so it could happen again someday. Since you say it's not critical for you, I'd like to keep trying to debug, but let me know if that changes and you want me to prioritize getting the object working without understanding what broke.
CThanks for the update! Yea no worries, take as much time as you need
Mis there/will there be a way to import the file from
db.dump() back into a database?VHow will workers figure out which D1 replica to read from? Something using the smart placement feature?
MWe'll share more once we get closer 
Msounds like you might want to debounce your searches every 300ms or so https://github.com/sindresorhus/debounce-fn
MIf you had 100,000 rows, each 200 bytes wide, and had to perform a full table scan every time - thatās approximately 5,000 read units out of the 25 billion per month included - or 5 million full table scans. Free.
MThis isnāt to say an index - https://developers.cloudflare.com/d1/learning/using-indexes/
- https://stackoverflow.com/a/57082926/556573
- isnāt useful, but unless you have millions of rows or really wide rows, cost isnāt likely to be a problem. Performance is the reason.
Indexes enable D1 to improve query performance over the indexed columns for common (popular) queries by reducing the amount of data (number of rows) ā¦
Stack Overflow
There is sort order when creating index for sqlite.
https://sqlite.org/lang_createindex.html
Each column name or expression can be followed by one of the "ASC" or "DESC" keywords to indicate ...
https://sqlite.org/lang_createindex.html
Each column name or expression can be followed by one of the "ASC" or "DESC" keywords to indicate ...
MIndexes that optimize ORDER BY must consider the ASC/DESC and NULLS FIRST/LAST modifiers.
MIād suggest following the D1 tutorial if you havenāt yet: https://developers.cloudflare.com/d1/get-started/
- this will show you how to query a D1 DB from a Worker.
NI remember seeing an intellij plugin for d1, but I can't find it anywhere, could somebody help me?
N
Mcc @geelen
MWhy indexes are useful (I was writing a little demo app w/ Remix...)
ORDER BY random() LIMIT 25-> 78ms query execution timeORDER BY created_at DESC LIMIT 25with an index oncreated_at DESC-> 0.20ms query execution time
JI just noticed a table called 
_cf_KV on every D1 database. This table has two columns, keyvalue
HThat might be prep for the SQL-on-DOs thing they are coming out with soon. According to Kenton(https://canary.discord.com/channels/595317990191398933/773219443911819284/1128025139632681131), regular DO KV storage will be stored in a SQLite Table, which might be what that is...
JThanks for the insights, that's very interesting and definitely good do know!
MNothing to do with KV the product
Mis there a size limit to what you can do with
db.batch() and/or db.exec()?
BIt appears
npx wrangler d1 migrations apply some-db --experimental-backend only works when logged in via npx wrangler login. In CI, however, where I'm not running npx wrangler login but instead providing a CLOUDFLARE_API_TOKEN(I've confirmed I can run commands such as
npx wrangler pages deploy dist --project-name abc in CI only providing the CLOUDFLARE_API_TOKENYeah, also provided that. Unfortunately I don't see any additional output.
BSince it works locally with
wrangler logini.e. could I somehow make
wrangler login