How we upgraded our 4TB Postgres database(retool.com) |
How we upgraded our 4TB Postgres database(retool.com) |
THERE IS NO CLOUD: It’s just someone else’s computer
The big names started using no-sql type stuff because their instances got 2-3 orders of magnitude larger, and that didn't work. It adds a lot of other overhead and problems doing all the denormalization though, but if you literally have multi-PB metadata stores, not like you have a choice.
Then everyone started copying them without knowing why.... and then everyone forgot how much you can actually do with a normal database.
And hardware has been getting better and cheaper, which makes it only more so.
Still not a good idea to store multi-PB metadata stores in a single DB though.
People tend to have a very bad sense of what constitutes large scale. It usually maps to "larger than the largest thing I've personally seen". So they hear "Use X instead of Y when operating at scale", and all of a sudden we have people implementing distributed datastore for a few MB of data.
Having gone downward in scale over the last few years of my career it has been eye opening how many people tell me X won't work due to "our scale", and I point out I have already used X in prior jobs for scale that's much larger than what we have.
There’s nothing inherently wrong with a big db instance. The cloud providers have fantastic automation around multi-az masters, read replicas and failover. They even do cross region or cross account replication.
That being said. Multi PB would raise an eyebrow.
Of course these systems have warm standbys, dedicated backup infrastructure and so it’s not really a “single machine”; but I’ve seen 80TiB Postgres instances back in 2011.
The biggest issue we have with these giant dbs is they require pretty massive amounts of RAM. That's currently our main bottle neck.
But I agree. While our design is pretty bad in a few ways, the amount of data that we are able to serve from these big DBs is impressive. We have something like 6 dedicated servers for a company with something like 300 apps. A hand full of them hit dedicated dbs.
Were I to redesign the system, I'd have more tiny dedicated dbs per app to avoid a lot of the noisy neighbor/scaling problems we've had. But at the same time, It's impressive how far this design has gotten us and appears to have a lot more legs on it.
When I come back to web 1.0 apps, I’m often surprised that it does a round-trip to the server in less than 200ms, and reloads the page seamlessly, including a full 5ms SQL query for 5k rows and returned them in the page (=a full 1MB of data, with basically no JS).
The other consideration that isn't initially obvious, is how you may hit an upper bound for resources in most modern environments. If your database is sitting on top of a virtual or containerized environment, your single instance database will be limited in resources (CPU/memory/network) to a single node of the cluster. You could also eventually hit the same problem on bare metal.
That said there are some very high density systems available. You may also not need the ability to scale as large as I am talking, or choose to shard and scale your database horizontally at later time.
If your project gets big enough you might also start wanting to replicate your data to localize it closer to the user. Another strategy might be to cache the data locally to the user.
There are positive and negatives with a single node or cluster. If retools database was clustered they would have been able to do a rolling upgrade though.
You can scale quite far vertically and avoid all the clustering headaches for a long time these days. With EPYCs you can get 128C/256T, 128PCIe lanes (= 32 4x NVMes = ~half a petabyte of low-latency storage, minus whatever you need for your network cards), 4TB of RAM in a single machine. Of course that'll cost you an arm and a leg and maybe a kidney too, but so would renting the equivalent in the cloud.
Not saying a single giant server won't work, but it does come with it's own set of very difficult-to-solve-once-you-build-it problems.
Sure, that's not the same thing as pure zero-downtime but for many applications it's OK to put the entire thing into read-only mode for a few minutes at a well selected time of day.
While it's in read-only mode (so no writes are being accepted) you can spin up a brand new DB server, upgrade it, finish copying data across - do all kinds of big changes. Then you switch read-only mode back off again when you're finished.
I've even worked with a team used this trick to migrate between two data centers without visible end-user downtime.
A trick I've always wanted to try for smaller changes is the ability to "pause" traffic at a load balancer - effectively to have a 5 second period where each incoming HTTP request appears to take 5 seconds longer to return, but actually it's being held by the load balancer until some underlying upgrade has completed.
Depends how much you can get done in 5 seconds though!
For PostgresQL or MySQL/MariaDB, your options are more limited. Here are two that come to mind, there may be more:
# The "Dual Writer" approach
1. Spin up a new database cluster on the new version. 2. Get all your data into it (including dual writes to both the old and new version). 3. Once you're confident that the new version is 100% up to date, switch to using it as your primary database. 4. Shut down the old cluster.
# The eventually consistent approach
1. Put a queue in front of each service for writes, where each service of your system has its own database. 2. When you need to upgrade the database, stop consuming from the queue, upgrade in place (bringing the DB down temporarily) and resume consumption once things are back online. 3. No service can directly read from another service's database. Eventually consistent caches/projections service reads during normal service operation and during the upgrade.
A system like this is more flexible, but suffers from stale reads or temporary service degradation.
1. Route all traffic through pgbouncer in transaction pooling mode.
2. Logically replicate from old to new.
For failover:
1. Ensure replication is not far behind.
2. Issue a PAUSE on pgbouncer.
3. Wait for replication to be fully caught up.
4. Update pgbouncer config to point to the new database.
5. Issue a RELOAD on pgbouncer.
6. Issue a RESUME on pgbouncer.
Zero downtime; < 2s additional latency for in-flight queries at time of op is possible (and I've done it at scale).
Doing an active no-downtime database migration basically involves having a coherent row-level merge policy (assuming you AT LEAST have a per-row last updated column), or other tricks. Or maybe you temporarily write cell-level timestamps and then drop it later.
Or if you have data that expires on a window, you just do double-writes for that period and then switch over.
I'm pretty sure that was the whole idea of RDBMS, to separate application from data. You badly lose the very moment when some of your data is in a different place -- on transactions, query planning, security, etc. -- so Codd thought "what if even different applications could use a single company-wide database?" Hence the "have everything in a single database" part should be the last compromise you're forced to make, not the first one.
The attitude “our company = scale and scale = nosql” is prevalent enough that even if you know better, it’s probably in your interest to play the game. It’s the one “scalability fact” everyone knows, and a shortcut to sounding smart in front of management when you can’t grasp or haven’t taken the time to dig in on the details.
NewSQL technology promises to make this more automated, which is definitely a good thing, but unless you are Google or have a use case that needs it, it probably isn't worth adopting it yet until they are more mature.
Stuff like “CRUD enterprise app. 1 large-ish Postgres node. 10k tenants. 100 tables with lots of foreign key lookups, 100gb on disk. Db is… kinda slow, and final web requests take ~1 sec.”
The toughest thing is knowing what is normal for multi tenant data with lots of relational info used (compared to more large and popular companies that tend to have relatively simple data models)
a lot of "kids these days" dont seem to learn that
by that I mean young folks born into this new world with endless cloud services and scaling-means-Google propaganda
a single modern server-class machine is essentially a supercomputer by 80s standards and too many folks are confused about just how much it can achieve if the software is written correctly
> We reasoned this was likely safe, as Retool’s product logic performs its own consistency checks, and also doesn’t delete from the referenced tables, meaning it was unlikely we’d be left with a dangling reference.
I was holding my breath here and I'm glad these were eventually turned back on.
Nobody should ever rely on their own product logic to ensure consistency of the database.
The database has features (constraints, transactions, etc) for this purpose which are guaranteed to work correctly and atomically in all situations such as database initiated rollbacks that your application will never have control over.
Happy to stick around and answer any questions you have.
It's written in Python, spins up a queue.Queue object, populates it with ranges of rows that need to be copied (based on min < ID < max ranges), starts up a bunch of Python threads and then each of those threads uses os.system() to run this:
psql "{source_url}" -c "COPY (SELECT * FROM ...) TO STDOUT" \
| psql "{dest_url}" -c "COPY {table_name} FROM STDIN"
psql -c "..." | pigz -c > file.tsv.gzOur database is less than 1TB. One master (for writes + some reads) + multiple slaves (read-only).
Here's what we did -
1, Launched a new read-only db with pg9.6, let's call it DB_A.
2, Stopped all offline tasks, and only maintained a minimal fleet of online servers (e.g., web, api...).
3, Changed all db hosts (no matter master or slave) in /etc/hosts on the minimal fleet of online servers (e.g., web, api...) to use old read-only db with pg9.6, let's call it DB_B. From this point on, all write ops should fail.
4, Ran pg_upgrade (with --link) on DB_A to upgrade to pg11, and promoted it to be a master db.
5, Changed /etc/hosts on the minimal fleet of online servers (e.g., web, api...) to use DB_A for all db hosts. By this point, DB_A is a master db. And write ops should be good now.
6, Changed /etc/hosts for all other servers and brought back all services.
Step 4 is the most critical. If it fails or runs too long (e.g., more than 10 minutes), then we had to rollback by changing /etc/hosts on those online servers.
We carefully rehearsed these steps for an entire week, and timed each step. By the time we did it on production, we knew how many seconds/minutes each step would take. And we tried to automate as many things as possible in bash scripts.
We missed one or two pieces of reconnecting things afterwards, and some of that seems to be limitations of Heroku Postgres that we couldn’t change. Hopefully those keep improving.
https://cloud.google.com/blog/products/databases/cloud-sql-l...
The biggest problem with all the cloud providers is that you won't know exactly when this 10 minute downtime window will start
I guess the only advantage here is that you don't have to do 9->10->11->12->13->14 like in the past and maybe that was one of the blockers Azure has. AWS allows to skip some major versions but 9->14 is not possible.
The benefit of an inplace upgrade for postgres is you don't have to spin up another server, restore from backup, and run pg_upgrade yourself.
On top of that Azure postgres (limited to pg11) has essentially deprecated in place of their v2 Flexible tier with no official migration path.
If you want to pay multi-thousands dollars a month for a database server, it's WAY cheaper just to slap a server with a ton of drives in colocation.
Yeah, this migration might have been somewhat painful, but it looks like it's a situation the company experiences every few years.
An expensive can be used that year where a capital expenses vest over years so only a portion can be applied that year.
=> select pg_size_pretty(pg_database_size ('**'));
pg_size_pretty
----------------
99 TB
(1 row)
(The re-analyze of tables took a day or so though)
It's really just a handful of core people who did most of the work, crafting it so thoughtfully over the years and it has such a huge impact on the world.
Doing huge part of it before postgresql was as popular as it is today, spending countless hours on making some great design choices and implementing them carefully.
It seems unlikely any of them will read that but I'm so deeply grateful to these people. It allowed so many things to flourish on top thanks to it being open source and free.
Does Postgres still make sense for append only tables or maybe elastic or other kind of database would be more suitable?
We actually migrated from vanilla Postgres to Aurora that way with minimal risk / downtime, it was a really smooth process.
On the other hand, self built databases maintained by professionals were ALOT cheaper.
We are talking here about million of dollars bills only for databases per month.
Self hosted solution did cost around 300k per month.
This included precompiled kernel patches, tweaks to postgres engine etc.
Overall the investment of a year of work of dbadmins will probably return itself by ten times if not more.
I only wish my Oracle updates were so simple and bug-free.
The idea that you don't get a seamless upgrade of the database itself with PostgreSQL is absurd to me. The part about "maximizing the amount of time this upgrade buys is" is only necessary because of how difficult PostgreSQL makes upgrades. We upgrade to every new version of SQL Server. It's not that big of a deal.
With every PostgreSQL blog article I read, I become more and more of an SQL Server fanboy. At this point it's full-blown. So many "serious business" PostgreSQL ops posts are just nothingburgers in the SQL Server world.
Very small companies might be able to do this on 'no load day' but from a pure business perspective, running your db twice is easier and way less risky.
You could have done this even without downtime by letting your connection proxy handling the switch.
Non distributed RDBMS is a great (yet underrated) choice. Thank you for the good writeup.
I was thinking you could have a much less delicate migration experience next time (some years from now). So you can go for a quicker parallel "dump and restore" migration.
For example:
- Client side sharding in the application layer: you could shard your customers' data across N smaller DB instances (consistent hashing on customer ID)
- Moving the append-only data somewhere else than postgres prior to the upgrade. You don't need RDBMS capabilities for that stuff anyway. Look at Elasticsearch, Clickhouse, or any DB oriented to time series data.
WDYT?
The first bullet point is on our radar for the near term. We have a very natural shard key in our schema (the customer ID), with AFAIK no relationships across that shard key. And once we start horizontally sharding, we can do cool things like putting your data in a shard geographically close to you, which will greatly increase app performance for our non US customers. Exciting stuff coming down the pike!
There could be all kinds of reasons why that is tho (if it is in fact the case).
I thought it was interesting that they upgraded 4 major version numbers in one go. I kept expecting to read something about version compatibility and configuration but was surprised there was none. Are major upgrades like this just less of an issue with Postgres in general?
I've run into version incompatibilities before, but it was my fault – they were expertly documented in the release notes and I just hadn't read them (or sufficiently tested the upgrade before the live performance of it).
Anyway, thanks for food for thought.
If you're using a hosted DB service, you're (probably) stuck in needing/wanting to rehearse using the hosted service (which is what the blog post describes).
If they were running the DB on 'regular server' cloud instances, it seems just as good to me to rehearse with other cloud server instances versus "standalone" servers.
audit events -> queue -> elastic -> blob storage
Then the cloud and DBaaS was invented.
One invaluable thing, though: our application was from the beginning designed to do 100% of all the reads from a read-only slave _if the slave was up to sync_ (which it was 95% of the time). We could also identify testers/developers in the application itself, so we had them using the upgraded slave for two weeks before the actual upgrade.
This made it possible for us to filter out problems in the application/DB-layer, which were few, which means that we probably did a minor version upgrade.
But upgrading by replication is something I can recommend.
Generally you just upgrade the replicas; then promote a replica to be the new primary; then upgrade the old primary and turn it into a replica.
The actual "upgrade" step is quite fast, since it doesn't actually need to iterate over your tables' row data.
At large scale, the painful part of major-version MySQL upgrades tends to be performance testing, but that's performed separately and prior to the actual upgrade process. Third-party tools (pt-upgrade, proxysql mirroring, etc) help a lot with this.
I've built some very high throughput Postgres backed systems in my years, and doing application side foreign key constraints (FKC) does have its benefits. Doing this client side will result in constraints that are usually, but not always in sync with data. However, this kind of almost-consistency lets you do much higher throughput queries. An FKC is a read on every write, for example, and does limit write throughput. Of course, this isn't ok for some workloads, and you do proper FKC in the DB, but if you don't need absolute consistency, you can make writes far cheaper.
Ideally, everything that needs to change when a row is deleted would be changed automatically and atomically using database-side constraints and triggers. In practice, applications often need to sync state with external services that the database knows nothing about, so I understand your concerns.
ON DELETE RESTRICT, on the other hand, will result in errors just like any other query error that you can handle in your application. Nothing happened, so there's nothing to be notified of.
I’ve seen certain large-table queries improve by 10x when foreign keys were added.
As mentioned downthread, we're doing some work soon to remove the audit table from the database, which will cut storage usage by over 50%.
This is barely past the initial release of Cockroach. It would have been kind of crazy for the Retool team to use an experimental db with a lack of history when building up their product (that was not dependent on experimental new features)
- physical restore to new cluster
- pg_upgrade the new cluster
- catch up on logical wal logs from old cluster
- failover to new cluster
- STONITH
I think the above was not open to them because of the limitations of their managed PG instance. I haven't used Azure but GCP managed SQL has loads of limitations. It seems very common and I think is a major (and undiscussed) drawback of these managed instance.
But the truth is that very few of the people who use PG want to hear that things are better in the MS SQL community for reasons of prejudice and as a result you're being downvoted unfairly for pointing out PGs relative backwardness here.
Here's a nice thing about PostgreSQL over SQL Server to satiate the fans: SQL Server is absurdly expensive to run in the cloud. I can't believe anyone uses RDS for SQL Server. Even in EC2 it's horrifically expensive. That's the main reason we have an on-prem cluster.
They created a replica database running the new version, then switched over to it. Not too dissimilar to what you describe, although more work since they started out with only a single instance without replication support.
All of this is exactly what I'm talking about. This blog post describes kind of a nightmare process for something that is trivial in SQL Server. They actually needed a third party product from Citus just to successfully upgrade PostgreSQL! Stunning.
Keep in mind that they’re upgrading from a database version that’s almost 6 years old. Postgres has improved a lot in the last 5 major versions since then.
Another thing here is that I’m pretty sure they could have just done the in-place upgrade and it would have been fine. I’ve run pg_upgrade myself for a bunch of major versions now and it’s easy and doesn’t require dumping / restoring anything. Maybe there’s something else going on that I’m missing though.
What setup are you running with sql server to have it automatically failover? Is it a multi master configuration or are the additional nodes just read replicas?
These days Postgres actually allows logical replication so your servers can be running different versions at the same time, which allows for much smoother upgrades (haven’t tried that myself yet, don’t quote me on it!)
SQL Server is designed to run in a Windows Server Failover Cluster; the SQL Server-side feature is called "Always On availability groups" in an HA configuration. It's a single-master arrangement, you can either have a fully passive secondary (that's what we do) or read-only replicas. The WSFC handles managing quorum, that's what causes the automatic failover as soon as >50% of the nodes are running the new version.
EDIT: I realized you're looking for the other PostgreSQL blog posts. Here's an example of two recent HN posts about PostgreSQL issues that I pulled out of my comment history. Both of these blog posts exist because PostgreSQL doesn't have query hints. SQL Server has them; I've dealt with issues like these blog posts describe but they have trivial fixes in the SQL Server world. Nothing to write a blog post about. I don't have a link handy regarding PostgreSQL's txn id wraparound problem, but SQL Server doesn't have that problem, either.
When your audit trail is in DB, you can pretty easily surface audit events to your customers. Who changed what when is just another feature. Capturing audit events is also usually pretty smooth.
The folks doing the blob storage route, you would not BELIEVE the complexity they have to spin up to expose very simple histories etc. This matters a LOT in some spaces (financial etc), less so in others.
In my RDBMS model, who changed this field when from what to what is a basic select. You can even shard by recordID or similar if you want to reduce table scans, good select of indexes etc can be a huge help as well. In most cases users don't mind a bit of latency on these queries.
We won't have billions of rows though, so once uploaded to secondary storage we'll just clear the blob field and set a "processed" flag.
This way we can find all the relevant keys for a given order, invoice etc quickly based on a partial key search in the database, and transparently fetch from either db directly or secondary storage as needed.
e.g. https://docs.microsoft.com/en-us/azure/azure-sql/database/le...
Under 1 minute of downtime was acceptable for us, so:
systemctl stop postgresql-11 && \
pg_upgrade --link ... && \
systemctl start postgresql-12
If a minute of downtime isn't acceptable, then it presumably isn't acceptable in case of unexpected hardware failure either, and you'd be using one of the commercial multi-master extensions.
I suspect the way I'll be setting it up is much the same as what you describe in your WSFC configuration (with a little more manual wrangling, no doubt).
The main barrier against adopting MS-SQL is just the pricing and that it's not open source. Another thing that PostgreSQL seems to do a lot better than MS-SQL is in the extensibility department, hence we have things like TimescaleDB, Citus, EdgeDB, and a whole lot more. I can't really find anything like that for MS-SQL, but perhaps I missed it?
When does it stop being experimental to you?
Avoiding FKs enables sharding, reduces write locking time, and greatly simplifies online schema changes. These are essential requirements at extreme scale. (and in response to GP's point, at least in MySQL, FK constraints strictly harm performance; although the underlying index improves performance, you can have that without the FK constraint)
As for bigints for time values: storing UTC unix timestamps in a bigint avoids some issues with unexpected timezone conversions and DST conversions, as well as (in older mysql versions) unwanted auto-update behavior for the first timestamp column in a table. This one tends to be more of a "reduce support burden on the database team" type of requirement -- the risk of datetime or timestamp issues goes up as the number of product engineers massively outnumbers the db engineers, or once you have product engineers in many different timezones, data centers in many timezones, etc.
Of course, there are major trade-offs with these decisions. And they may or may not make sense for your former company's size and situation. But in any case these restrictions are not particularly unusual for MySQL.
The way I've done this involves logical replication also: https://news.ycombinator.com/item?id=31087197
(Perhaps you do have such problems, I don't know where you work! But 99%+ of companies don't have such problems and never will.)
Not sure where you get your metrics, but I would say a more general rule would be that the more people work on an evolving product that includes code and schema changes, then the more you need db constraints to enforce what it means to have correct data.
If only 1 or 2 people are involved in a db that changes pretty infrequently then possibly in the long term you can get away with it.
But if you have a constantly evolving product which must carry new features, logic changes and additions to the schema, then I would say you definitely need db constraints - FK and column. It only takes a few different developers to decide that T,F,Y,N,TRUE,FALSE,YES,NO,Null,NULL,None all mean the same thing, and you've got a slowly evolving mess on your hands.
I've done something similar, although it wasn't about upgrading the database. We needed to not only migrate data between different DB instances, but also between completely different data models (as part of refactoring). We had several options, such as proper replication + schema migration in the target DB, or by making the app itself write to two models at the same time (which would require a multi-stage release). It all sounded overly complex to me and prone to error, due to a lot of asynchronous code/queues running in parallel. I should also mention that our DB is sharded per tenant (i.e. per an organization). What I came up with was much simpler: I wrote a simple script which simply marked a shard read-only (for this feature), transformed and copied data via a simple HTTP interface, then marked it read-write again, and proceeded to the next shard. All other shards were read-write at a given moment. Since the migration window only affected a single shard at any given moment, no one noticed anything: for a tenant, it translated to 1-2 seconds of not being able to save. In case of problems it would also be easier to revert a few shards than the entire database.
I'm picturing your migration script looping over shards. It flips it to read-only, migrates, then flips back to read-write.
How did the app handle having some shards in read-write mode pre-migration and other shards in read-write post-migration simultaneously.
The app handled it by proxying calls to the new implementation if the shard was marked as "post-migration", the API stayed the same. If it was "in migration", all write operations returned an error. If the state was "pre-migration", it worked as before.
I don't already remember the details but it was something about the event queue or the notification queue which made me prefer this approach over the others. When a shard was in migration, queue processing was also temporarily halted.
Knowing that a shard is completely "frozen" during migration made it much easier to reason about the whole process.
But agreed, a daily dump to something can go a long way to unlocking other tools - in govt this is especially true not because the DBMS is hard to use, but because so many layers of consultants and others in the way it's not usable.
Sql server data is stored as a BTree structure. So a 10 -> 100gb growth ends up being roughly a 1/2 query performance slowdown (since it grows by a factor of log n) assuming good indexes are in place.
Filtered indexes can work pretty well for improving query performance. But ultimately we do have some tables which are either archived if we can or partitioned if we can't. SQL Server native partitioning is rough if the query patterns are all over the board.
The other thing that has helped is we've done a bit of application data shuffling. Moving heavy hitters onto new database servers that aren't as highly utilized.
We are currently in the process of getting read only replicas (always on) setup and configured in our applications. That will allow for a lot more load distribution.
EVENTUALLY, yes even read query performance also would degrade, but typically the insert / update load on a typical index is the first limiter.
The trick is to have the right indexes (which includes the interior structures of the storage data structure) so that queries jump quickly to the relevant data and ignore the rest. Like opening a book at the right page because the page number is known. Sometimes a close guess is good enough.
In addition, small indexes and tree interior nodes should stay hot in RAM between queries.
When the indexes are too large to fit in RAM, those get queried from storage as well, and at a low level it's analogous to the system finding the right page in an "index book", using an "index index" to get that page number. As many levels deep as you need. The number of levels is generally small.
For example, the following is something I worked on recently. It's a custom database (written by me) not Postgres, so the performance is higher but the table scaling principles are similar. The thing has 200GB tables at the moment, and when it's warmed up, querying a single value takes just one 4k read from disk, a single large sector, because the tree index fits comfortably in RAM.
It runs at approximately 1.1 million random-access queries/second from a single SSD on my machine, which is just a $110/month x86 server. The CPU has to work quite hard to keep up because the data is compressed, albeit with special query-friendly compression.
If there was very little RAM so nothing could be kept in it, the speed would drop by a factor of about 5, to 0.2 million queries/second. That shows you don't need a lot of RAM, it just helps.
Keeping the RAM and increasing table size to roughly 10TB the speed would drop by half to 0.5 million queries/second. In principle, with the same storage algorithms a table size of roughly 1000TB (1PB) would drop it to 0.3 million queries/second, and roughly 50,000TB (50PB) would drop it to 0.2 million. (But of course those sizes won't fit on a single SSD. A real system of that size would have more parallel components, and could have higher query performance.) You can grow to very large tables without much slowdown.
This is a bit of a simplification, but probably less than you'd think. It's definitely true in spirit - the assumptions that I'm making are pretty reasonable. Lots of people don't quite get their head around all this at first, but it's easier to understand with experience. It doesn't help that most pictures of B-Tree indexes are very misleading. It's closer to a bush than to a tree, really.
Proper indexing is key of course, and metrics to find bottlenecks.
Of course, if your application rarely makes use of older rows, it could still make sense to offload them to some kind of colder, cheaper storage.
As great as it can be to enforce rules within the database, a lot of them usually end up needing to be enforced at the application layer instead. Especially when performance at scale comes into play.
My goal is always to verify what I can in the database to minimize potential data cleanup. In my experience, app only verification always leads to future time investments to clean up the mess.
For a crude example, it's trivial for DB constraints verify (via a foreign key constraint) that all your contracts belong to some customer, but very difficult for DB constraints to verify that all your currently active contracts belong to a currently active customer, even if the definition of 'active' is some relatively simple business logic.
So in my experience it's not that rare to have some code-based data integrity tests that run various sanity checks on production data to verify things that DB constraints can not.
Applies to vendors, too. Had some data in Rackspace Files where "list files" would say X existed, but "GET X" got you a 404. Had an AWS RDS instance; query on it returned no results. Adding the "don't use the index" index hint caused it to return data. (Allegedly, this bug was fixed, but we had migrated off by that point, so I never got to confirm it.)
Conversely, I do like DB constraints, because if the DB constraint doesn't exist, then I guarantee you the production DB has a row that is a counter-example to whatever constraint you think the data should obey…
Well yes, Rackspace Files (aka OpenStack Swift) is eventually consistent. It says so literally in the first sentence of the documentation [1]. But this discussion is about relational databases with ACID guarantees, where the C is literally "consistent".
Out of pure curiosity, what are you replacing it with (if anything)? Just a simple rotating log file?
[1] https://github.com/icheishvili/audit-trigger [2] https://github.com/eulerto/wal2json
Using compression? Second instance for old data? Creating in-between states?
Just curious :)
The current application is Ethereum L1 state and state history, but it has useful properties for other applications. It's particularly good at being small and fast, and compressing time-varying blockchain-like or graph data.
As it's a prototype I'm not committing to final figures, but measurement, theory and prototype tests project the method to be significantly smaller and faster than other implementations, or at least competitive with the state of the art being researched by other groups.
> Why did you reinvent the wheel?
Different kind of wheel. No storage engine that I'm aware of has the desired combination of properties to get the size (small) and speed (IOPS, lower read & write amplification) in each of the types of operations required. Size and I/O are major bottlenecks for this type of application; in a way it's one of the worst cases for any kind of database or schema.
It's neither a B-tree nor an LSM-tree, (not a fractal tree either), because all of those are algorithmically poor for some of the operations required. I found another structure after being willing to "go there" relating the application to low-level storage behaviour, and reading older academic papers.
These data structures are not hard to understand or implement, once you get used to them. As I've been working on and off for many years on storage structures as a hobby (yeah, it's fun!), it's only natural to consider it an option when faced with an unusual performance challenge.
It also allowed me to leverage separate work I've done on raw Linux I/O performance (for filesystems, VMs etc), which is how random-access reads are able to reach millions/s on a single NVMe SSD.
> Is it as durable as Postgres?
Yes.
Modulo implementation bugs (because it won't have the scrutiny and many eyes/years of testing that Postgres does).
> Any links?
Not at this time, sorry!
Sometimes making 10+ million dollar decisions off that gut feel with literally zero data on what is actually going on.
It rarely works out well, but hey, have to leave that opening for competition somehow I guess?
And I'm not talking about 'why didn't they spend 6 months optimizing that one call which would save them $50 type stuff'. I mean literally zero idea what is going on, what actual performance issues are, etc.
[1] https://www.amazon.com/SanDisk-Extreme-microSDXC-Memory-Adap...
It seems like the last 4-5 years though have really made it super common again. Bubble maybe?
Huge horde of newbs?
Maybe I’m getting crustier.
I remember it was SUPER bad before the dot-com crash, all the fake it ‘til you make it too. I even had someone claim 10 years of Java experience who couldn’t write out a basic class on a whiteboard at all, and tons of folks starting that literally couldn’t write a hello world in the language they claimed experience in, and this was before decent GUI IDEs.
Cloud providers have successfully redefined the baseline performance of a server in the minds of a lot of developers. Many people don't understand just how powerful (and at the same time cheap) a single physical machine can be when all they've used is shitty overpriced AWS instances, so no wonder they have no confidence in putting a standard RDBMS in there when anything above 4GB of RAM will cost you an arm and a leg, therefore they're looking for "magic" workarounds, which the business often accepts - it's easier to get them to pay lots of $$$$ for running a "web-scale" DB than paying the same amount for a Postgres instance, or God forbid, actually opting for a bare-metal server outside of the cloud.
In my career I've seen significant amount of time & effort being wasted on workarounds such as deferring very trivial tasks onto queues or building an insanely-distributed system where the proper solution would've been to throw more hardware at it (even expensive AWS instances would've been cost-effective if you count the amount of developer time spent working around the problem).
I feel like sometimes the pendulum has swung too far the other way, where people deny that there ARE people dealing with actual scale problems.
I think it is the same frustration I get when I call my ISP for tech support and they tell me to reboot my computer. I realize that they are giving advice for the average person, but it sucks having to sit through it.
I feel similar frustrations with commenters saying I am doing it wrong by not moving everything to the cloud… I work for a CDN, we would go out of business pretty quickly if we moved everything to the cloud. Oh well.
Now when dealing with someone convinced that their single TB of data is google scale, the harder issue is changing that belief. But at least you know where they stand.
One gotcha here is not all PBs are equal. My field also is a case where multi-PB datastores are common. However for the most part those data sit at rest in S3 or similar. They'll occasionally be pulled in chunks of a couple TB at most. But when you talk to people they'll flash their "do you know how big our storage budget is?" badge at the drop of a hat. It gets used to explain all sorts of compute patterns. Meanwhile, all they need is a large S3 footprint and a machine with a reasonable amount of RAM.
My new fav rant is the voice phone systems for Kaiser, which makes me say 'Yes or No' constantly - but literally can only hear me somehow if I'm yelling. And they don't tell you to press a number to say yes or no until after you've failed several times with the voice system.
All human convos have zero issues, not even a little faint.