Parallel Commits: A New Atomic Commit Protocol for Distributed Transactions
1–10 of 41 posts
Re: Parallel Commits: A New Atomic Commit Protocol for Distributed Transactions
#2Why is it so important for transactions to be interactive? I feel like this is not a feature that I use frequently as an application developer and it makes distributed transactions so much harder.
It seems like something that accidentally came along from SQL, and is imposing a large cost.
Re: Parallel Commits: A New Atomic Commit Protocol for Distributed Transactions
#3Why is it so important for transactions to be interactive? I feel like this is not a feature that I use frequently as an application developer and it makes distributed transactions so much harder. It seems like something that accidentally came along from SQL, and is imposing a large cost.
Interactive may be a little misleading here. Consider a common case in any java application at my current place of employment.
1. A controller starts a transaction.
2. Some code looks up a record in a database.
3. Based on the results of that lookup we run 1 of two possible writes to a table in that database.
Knowing what kind of write you will do in the transaction requires knowing the result of the lookup which itself must be run in the transaction. Now take this small case and imagine it in the real world scenario where there might many such lookup specific write logic all wrapped in a single transaction and the space of possible combinations of statements you will want to execute in the transaction is large. None of this is interactive in the sense of someone working in a shell. But it is interactive in a way that makes the transaction useful in the real word.
Re: Parallel Commits: A New Atomic Commit Protocol for Distributed Transactions
#4Why is it so important for transactions to be interactive? I feel like this is not a feature that I use frequently as an application developer and it makes distributed transactions so much harder. It seems like something that accidentally came along from SQL, and is imposing a large cost.
This article implies that it reduces latency.
Re: Parallel Commits: A New Atomic Commit Protocol for Distributed Transactions
#5Why is it so important for transactions to be interactive? I feel like this is not a feature that I use frequently as an application developer and it makes distributed transactions so much harder. It seems like something that accidentally came along from SQL, and is imposing a large cost.
Interactive may be a little misleading here. Consider a common case in any java application at my current place of employment. 1. A controller starts a transaction. 2. Some code looks up a record in a database. 3. Based on the results of that lookup we run 1 of two possible writes to a table in that database. Knowing what kind of write you will do in the transaction requires knowing the result of the lookup which its…
Exactly. What Calvin (and FaunaDB, I think) does is allow you to run an arbitrary function which takes input on the database. That function can do whatever it wants, including any number of reads, writes, and arbitrary logic based on the reads. But critically, it can't talk back to the calling client, except for to send the final result.
This allows you to implement the pattern you describe, which I agree is common, but in a dramatically simpler way.
Having the database, which is not really a single thing, but a swarm of computers spread across the globe separated by unreliable links who are trying to stay in consensus, pause their work to hear back from the client just seems ... well it seems miraculous that it can work at all.
The Calvin way is so much easier, it seems like there must be some very good reason that it's not what CockroachDB does. But I've never heard what that reason is.
Re: Parallel Commits: A New Atomic Commit Protocol for Distributed Transactions
#6I like the elegance and simplicity of two-phase commits. I didn't understand the criticism in the article; maybe it's something specific to CockroachDB. In my experience with two-phase commits, if the system crashes before a transaction is fully processed and committed, it should be fully reprocessed (from scratch) once the server restarts.
In one of my open source projects with RethinkDB (which doesn't natively support atomic transactions), I implemented a distributed 'parallel' two-phase commit mechanism by assigning each pending record a shard key (an integer derived from a hash of the record id); then worker servers would decide which subset/range of DB records to process/commit based on a hash of their own server id (which would tell them which range of shard keys/records they were responsible for). Only when a record had been fully processed, its status would be updated as committed by writing a 'settled' flag on that record.
Whenever a server failed and restarted, it would pick up processing from the last successful commit. If a server did not restart, the worker count would be updated and remaining workers would redistribute the partitions among themselves based on the shard keys of the records.
Re: Parallel Commits: A New Atomic Commit Protocol for Distributed Transactions
#7Earlier quoted context omitted.
Interactive may be a little misleading here. Consider a common case in any java application at my current place of employment. 1. A controller starts a transaction. 2. Some code looks up a record in a database. 3. Based on the results of that lookup we run 1 of two possible writes to a table in that database. Knowing what kind of write you will do in the transaction requires knowing the result of the lookup which its…
Exactly. What Calvin (and FaunaDB, I think) does is allow you to run an arbitrary function which takes input on the database. That function can do whatever it wants, including any number of reads, writes, and arbitrary logic based on the reads. But critically, it can't talk back to the calling client, except for to send the final result. This allows you to implement the pattern you describe, which I agree is common,…
Wouldn't you have to implement a DSL and all the parts related to it for that to work? Also things like serialization cost. What if I have a large local in memory structure I want to base my query results off of? Funny enough this is kind of similar to the problem solved by things like apache beam.
Re: Parallel Commits: A New Atomic Commit Protocol for Distributed Transactions
#8Earlier quoted context omitted.
Exactly. What Calvin (and FaunaDB, I think) does is allow you to run an arbitrary function which takes input on the database. That function can do whatever it wants, including any number of reads, writes, and arbitrary logic based on the reads. But critically, it can't talk back to the calling client, except for to send the final result. This allows you to implement the pattern you describe, which I agree is common,…
Wouldn't you have to implement a DSL and all the parts related to it for that to work? Also things like serialization cost. What if I have a large local in memory structure I want to base my query results off of? Funny enough this is kind of similar to the problem solved by things like apache beam.
No you'd need a complete programming language. But SQL is basically one already, most variants are already turing-complete.
You'd also have to provide the programming environment with concepts of cursors and so on so they could page through data efficiently.
Re: Parallel Commits: A New Atomic Commit Protocol for Distributed Transactions
#9Wow, this is impressive.
Re: Parallel Commits: A New Atomic Commit Protocol for Distributed Transactions
#10I like the elegance and simplicity of two-phase commits. I didn't understand the criticism in the article; maybe it's something specific to CockroachDB. In my experience with two-phase commits, if the system crashes before a transaction is fully processed and committed, it should be fully reprocessed (from scratch) once the server restarts. In one of my open source projects with RethinkDB (which doesn't natively supp…
The "criticism" as it pertains to CockroachDB and off-the-shelf 2PC is less so about 2PC in isolation and more so about layering 2PC on top of consensus groups used to persist records. When any given txn does the "prepare" phase, it lays down markers for a possible upcoming commit. If the 2PC coordinator fails in an inopportune moment, there's a delay between the failure and the markers being cleaned up (whether or not the transaction is "reprocessed" or aborted). The reason why this delay is problematic is because any subsequent transactions that happen upon said markers, they just have to wait for the resolution ("commit"/"aborted") aka it blocks. So clearly recovery must be built into 2PC, i.e. the transaction state itself must be persisted. This is done so in the same way the markers/regular writes are, through consensus. But marking the transaction state as "committed" can only happen once we're guaranteed that all the individual write markers are persisted. Which adds a second round of consensus.