Hi Jonathan, I will summarise my position below, that I have outlined at various points in the other thread, and then I would be interested to hear how you propose we move forwards. I will commit to responding the same day to any email I receive before 7pm GMT, and to engaging with each of your points. I would appreciate it if you could make similar commitments so that we may conclude this discussion in a reasonable time frame and conduct a vote on CEP-15.
I also reiterate my standing invitation to an open video chat, to discuss anything you like, for as long as you like. Please nominate a suitable time and day. ==TL;DR== CEP-15 does not narrow our future options, it only broadens them. Accord is a distributed consensus protocol, so these techniques may build upon it without penalty. Alternatively, these approaches may simply live alongside Accord. Since these alternative approaches do not achieve the goals of the CEP, and this CEP only enhances your ability to pursue them, it seems hard to conclude it should not proceed. ==Goals== Our goals are first order principles: we want strict serializable cross-shard isolation that is highly available and can be scaled while maintaining optimal and predictable latency. Anything less, and the CEP is not achieved. As outlined already (except SLOG, which I address below), these alternative approaches do not achieve these goals. ==Compatibility with other approaches== 0. In general, research systems are not irreducible - they are an assembly of ideas that can be mixed together. Accord is a distributed consensus protocol. These other protocols may utilise it without penalty for consensus, in many cases obtaining improved characteristics. Conversely, Accord may itself directly integrate some of these ideas. 1. Cockroach, YugaByte, Dynamo et al utilize read and write intents, the same as outlined as a technique for interactive transactions with Accord. They manage these in a distributed state machine with per-shard consensus, permitting them to achieve serializable isolation. This same technique can be used with Accord, with the advantage that strict serializable isolation would be achievable. For simple transactions we would be able to execute with “pure” Accord and retain its execution advantage. Accord does not disadvantage this approach, it is only enhanced and made easier. 2. Calvin: Accord is broadly functionally equivalent, only leaderless, thereby achieving better global latency properties. 3. SLOG: This is essentially Calvin. The main modification is that we may assign data a home region, so that transactions may be faster if they participate in just one region, and slower if they involve multiple regions. Note that this protocol does not achieve global serializability without either losing consistency or availability under network partition or paying a WAN cost. In its consistent mode SLOG therefore remains slower than Accord for both single-home and multi-home transactions. Accord requires one WAN penalty for linearizing a transaction (competing transactions pay this cost simultaneously, as with SLOG), however this is achieved for global clients, whereas SLOG must cross the WAN multiple times for transactions initiated from outside their home, and for all multi-home transactions. As discussed elsewhere, a future optimisation with Accord is to temporarily “home” competing transaction for execution only, so that there is no additional WAN penalty when executing competing transactions. This would confer the same performance advantages as SLOG, without any of its penalties for multi-home transactions or heterogenous latency characteristics, nor any of the complexities of re-homing data, thus avoiding these unpredictable performance characteristics. For those use cases that do not require high availability, it would be possible to implement a “home” region setup with Accord, as with SLOG. This is not an idea that is exclusive to this particular system. We even discussed this briefly in the call, as some use cases do indeed prefer this trade-off. SLOG additionally offers a kind of “home group” multi-home optimisation for clusters with many regions, that accept availability loss if fewer than half of their regions fail (e.g. in the paper 6 regions in pairs of 2 for availability). This is also exploitable by Accord, and something we can pursue as a future optimisation, as users explore such topologies in the real world. ==Responding to specific points== >because it was asserted in the CEP-15 thread that Accord could support SQL by >applying known techniques on top. This is mistaken. Deterministic systems like >Calvin or SLOG or Accord can support queries where the rows affected are not >known in advance using a technique that Abadi calls OLLP Language is hard and it is easy to conflate things. Here you seem to be discussing abort-free interactive transactions, not SQL. SQL does not necessitate support for interactive transactions, let alone abort-free ones. The technique you mention can support SQL scripts, and also interactive client transactions that may be aborted by the server. However, see [1] which may support all of these properties. From: Blake Eggleston <beggles...@apple.com.INVALID> Date: Sunday, 10 October 2021 at 05:17 To: dev@cassandra.apache.org <dev@cassandra.apache.org> Subject: Re: Tradeoffs for Cassandra transaction management 1. Is it worth giving up local latencies to get full global consistency? Most LWT use cases use LOCAL_SERIAL. This isn’t a tradeoff that needs to be made. There’s nothing about Accord that prevents performing consensus in one DC and replicating the writes to others. That’s not in scope for the initial work, but there’s no reason it couldn’t be handled as a follow on if needed. I agree with Jeff that LOCAL_SERIAL and LWTs are not usually done with a full understanding of the implications, but there are some valid use cases. For instance, you can enable an OLAP service to operate against another DC without impacting the primary, assuming the service can tolerate inconsistency for data written since the last repair, and there are some others. 2. Is it worth giving up the possibility of SQL support, to get the benefits of deterministic transaction design? This is a false dilemma. Today, we’re proposing a deterministic transaction design that addresses some very common user pain points. SQL addresses different user pain point. If someone wants to add an sql implementation in the future they can a) build it on top of accord b) extend or improve accord or c) implement a separate system. The right choice will depend on their goals, but accord won’t prevent work on it, the same way the original lwt design isn’t preventing work on multi-partition transactions. In the worst case, if the goals of a hypothetical sql project are different enough to make them incompatible with accord, I don’t see any reason why we couldn’t have 2 separate consensus systems, so long as people are willing to maintain them and the use cases and available technologies justify it. -Blake > On Oct 9, 2021, at 9:54 AM, Jonathan Ellis <jbel...@gmail.com> wrote: > > * Hi all,After calling several times for a broader discussion of goals and > tradeoffs around transaction management in the CEP-15 thread, I’ve put > together a short analysis to kick that off.Here is a table that summarizes > the state of the art for distributed transactions that offer > serializability, i.e., a superset of what you can get with LWT. (The most > interesting option that this eliminates is RAMP.)Since I'm not sure how > this will render outside gmail, I've also uploaded it here: > https://imgur.com/a/SCZ8jex > <https://imgur.com/a/SCZ8jex>SpannerCockroachCalvin/FaunaSLOG (see > below)Write latencyGlobal Paxos, plus 2pc for multi-partition.For > intercontinental replication this is 100+ms. Cloud Spanner does not allow > truly global deployments for this reason.Single-region Paxos, plus 2pc. > I’m not very clear on how this works but it results in non-strict > serializability.I didn’t find actual numbers for CR other than “2ms in a > single AZ” which is not a typical scenario.Global Raft. Fauna posts actual > numbers of ~70ms in production which I assume corresponds to a multi-region > deployment with all regions in the USA. SLOG paper says true global Calvin > is 200+ms.Single-region Paxos (common case) with fallback to multi-region > Paxos.Under 10ms.Scalability bottlenecksLocks held during cross-region > replicationSame as SpannerOLLP approach required when PKs are not known in > advance (mostly for indexed queries) -- results in retries under > contentionSame as CalvinRead latency at serial consistencyTimestamp from > Paxos leader (may be cross-region), then read from local replica.Same as > Spanner, I thinkSame as writesSame as writesMaximum serializability > flavorStrictUn-strictStrictStrictSupport for other isolation > levels?SnapshotNoSnapshot (in Fauna)Paper mentions dropping from > strict-serializable to only serializable. Probably could also support > Snapshot like Fauna.Interactive transaction support (req’d for > SQL)YesYesNoNoPotential for grafting onto C*NightmareNightmareReasonable, > Calvin is relatively simple and the storage assumptions it makes are > minimalI haven’t thought about this enough. SLOG may require versioned > storage, e.g. see this comment > <http://dbmsmusings.blogspot.com/2019/10/introducing-slog-cheating-low-latency.html?showComment=1570497003296#c5976719429355924873>.(I > have not included Accord here because it’s not sufficiently clear to me how > to create a full transaction manager from the Accord protocol, so I can’t > analyze many of the properties such a system would have. The most obvious > solution would be “Calvin but with Accord instead of Raft”, but since > Accord already does some Calvin-like things that seems like it would result > in some suboptimal redundancy.)After putting the above together it seems to > me that the two main areas of tradeoff are, 1. Is it worth giving up local > latencies to get full global consistency? Most LWT use cases use > LOCAL_SERIAL. While all of the above have more efficient designs than LWT, > it’s still true that global serialization will require 100+ms in the > general case due to physical transmission latency. So a design that allows > local serialization with EC between regions, or a design (like SLOG) that > automatically infers a “home” region that can do local consensus in the > common case without giving up global serializability, is desirable.2. Is it > worth giving up the possibility of SQL support, to get the benefits of > deterministic transaction design? To be clear, these benefits include very > significant ones around simplicity of design, higher write throughput, and > (in SLOG) lower read and write latencies.I’ll doubleclick on #2 because it > was asserted in the CEP-15 thread that Accord could support SQL by applying > known techniques on top. This is mistaken. Deterministic systems like > Calvin or SLOG or Accord can support queries where the rows affected are > not known in advance using a technique that Abadi calls OLLP (Optimistic > Lock Location Prediction), but this does not help when the transaction > logic is not known in advance.Here is Daniel Abadi’s explanation of OLLP > from “An Overview of Deterministic Database Systems > <https://cacm.acm.org/magazines/2018/9/230601-an-overview-of-deterministic-database-systems/fulltext?mobile=false>:”In > practice, deterministic database systems that use ordered locking do not > wait until runtime for transactions to determine their access-sets. > Instead, they use a technique called OLLP where if a transaction does not > know its access-sets in advance, it is not inserted into the input log. > Instead, it is run in a trial mode that does not write to the database > state, but determines what it would have read or written to if it was > actually being processed. It is then annotated with the access-sets > determined during the trial run, and submitted to the input log for actual > processing. In the actual run, every replica processes the transaction > deterministically, acquiring locks for the transaction based on the > estimate from the trial run. In some cases, database state may have changed > in a way that the access sets estimates are now incorrect. Since a > transaction cannot read or write data for which it does not have a lock, it > must abort as soon as it realizes that it acquired the wrong set of locks. > But since the transaction is being processed deterministically at this > point, every replica will independently come to the same conclusion that > the wrong set of locks were acquired, and will all independently decide to > abort the transaction. The transaction then gets resubmitted to the input > log with the new access-set estimates annotated.Clearly this does not work > if the server-visible logic changes between runs. For instance, consider > this simple interactive transaction:cursor.execute("BEGIN > TRANSACTION")count = cursor.execute("SELECT count FROM inventory WHERE id = > 1").result[0]if count > 0: cursor.execute("UPDATE inventory SET count = > count - 1 WHERE id = 1")cursor.execute("COMMIT TRANSACTION")The first > problem is that it’s far from clear how to do a “trial run” of a > transaction that the server only knows pieces of at a time. But even > worse, the server only knows that it got either a SELECT, or a SELECT > followed by an UPDATE. It doesn’t know anything about the logic that would > drive a change in those statements. So if the value read changes between > trial run and execution, there is no possibility of transparently retrying, > you’re just screwed and have to report failure.So Abadi concludes,[A]ll > recent [deterministic database] implementations have limited or no support > for interactive transactions, thereby preventing their use in many existing > deployments. If the advantages of deterministic database systems will be > realized in the coming years, one of two things must occur: either database > users must accept a stored procedure interface to the system [instead of > client-side SQL], or additional research must be performed in order to > enable improved support for interactive transactions.TLDR:We need to decide > if we want to give users local transaction latencies, either with an > approach inspired by SLOG or with tuneable serializability like LWT > (trading away global consistency). I think the answer here is clearly Yes, > we have abundant evidence from LWT that people care a great deal about > latency, and specifically that they are willing to live with > cross-datacenter eventual consistency to get low local latencies.We also > need to decide if we eventually want to support full SQL. I think this one > is less clear, there are strong arguments both ways.P.S. SLOG deserves more > attention. Here are links to the paper > <http://www.vldb.org/pvldb/vol12/p1747-ren.pdf>, Abadi’s writeup > <http://dbmsmusings.blogspot.com/2019/10/introducing-slog-cheating-low-latency.html>, > and Murat Demirbas’s reading group compares SLOG to something called Ocean > Vista that I’ve never heard of but which reminds me of Accord > <http://muratbuffalo.blogspot.com/2020/11/ocean-vista-gossip-based-visibility.html>.* > -- > Jonathan Ellis > co-founder, http://www.datastax.com > @spyced --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@cassandra.apache.org For additional commands, e-mail: dev-h...@cassandra.apache.org