Pro tip: If you go to touch.facebook.com and request a desktop site, it will let you access your messages on a mobile web browser without installing messenger. :D
I feel like they should have done a better job of defining quality of life. A person who is born and grows old and dies in North Dakota does not have a better life than a person who does the same in California. If only for the education, career opportunities, weather, culture, and earning potential. Yea there are problems, but there are enormous benefits too.
Neanderthalensis, like Sapien, is a species name. When we talk about these things in the context of evolutionary biology, those precise terms tend to work better than a more broad and abstract word like "Humans."
But, in articles like this, you see mixed jargons. One academic source may specialize in evolutionary biology and its particular jargon, and another source may specialize in anthropology. Their disparate perspectives and jargons are filtered through the lens of a journalist who specializes in neither field.
Each group assigns a different meaning to these words.
There is a precise and correct word in the academic context for each category of early hominids, but it probably isn't realistic to expect writers to use jargon from a field other than the one they specialize in.
I did read the article, but I can't find any mention of addressing the "Write amplification" issue as described by Uber when they moved away from postgres. https://eng.uber.com/mysql-migration/ I had heard talk on Software Engineering Daily that this new major revision was supposed to address that.
Is this issue resolved by the new "Logical replication" feature? It doesn't seem directly related, but it seems like maybe that is what he is referring to in this blog post?
There's a patch reducing write amplifications (when caused by indexes), by a significant degree. Unfortunately it didn't quite get ready in time for the feature freeze of 10 - as it affects the on-disk format, we considered the risk to be too high.
As the author of the patch I don't quite agree to it. But it's true that the patch did not receive adequate review even though most of the on-disk changes were known and coded at least 7 months before the feature freeze. So it's hard to tell which part of the patch wasn't ready. But there is always next cycle. So lets work towards getting it ready for v11.
Write amplification is a result of PostgreSQL's decision to not used clustered indexes, there's not much that can be done to avoid it without a massive redesign of the storage engine - though there are patches out there to reduce the penalty in some cases. In all reality though, Uber wanted a key-value store and not an RDBMS, MySQL was a better choice for this since InnoDB isn't much more than a fast K/V store (hence why MySQL uses clustered indexes).
> Write amplification is a result of PostgreSQL's decision to not used clustered indexes, there's not much that can be done to avoid it without a massive redesign of the storage engine
I don't think that's entirely accurate - the issue is more that indexes contain pointers to the heap position (simplified) of a tuple, rather than being indirect and pointing to the primary key, which then is resolved by another index (be that clustered / primary or not).
Updates already don't have to update indexes iff none of the indexed columns change (HOT - Heap-Only-Tuples). The proposed change (WARM - write amplification reduction method), allows to avoid updating indexes on non-changing columns, even if other indexes change.
Not arguing with your assessment of Uber's requirements; but in general, why do you view InnoDB as not much more than a K/V store? And why do you equate clustered indexes with K/V stores?
InnoDB is a complex piece of software, supporting transactions, row-level locking, MVCC, schemas, secondary indexes, crash recovery, hot copy/backup, complex caching and buffering, many tunables, and extensive metrics visibility. Just because it's more appropriate for Uber's rather unusual EAV-like use-case, this doesn't mean InnoDB is a glorified K/V store.
Re: clustered indexes, it's a storage engine architecture choice with well-known trade-offs, both positive and negative. SQL Server also uses clustered indexes and is widely respected among database experts.
Regarding the topic overall, there are use-cases where Postgres is the best choice, and there are use-cases where it isn't. That doesn't inherently mean that other databases are uniformly worse. People like to trash MySQL, sometimes for completely valid reasons, but other times for FUD. But fwiw, several of the major features in Postgres 10 have already been supported in MySQL/InnoDB for a long time, in some cases for over a decade. Of course, that goes both ways; there are awesome major features that Postgres has had for a decade that MySQL still lacks.
Clustered indexes aren't an all-or-nothing choice. SQL Server allows heap tables without any clustered indexes, tables with a clustered index on the primary key, and tables clustered on a secondary index with the primary key as a non-clustered index. It is really nice to have all three of those options available.
Have the Postgres thought about adding support for more than one storage engine? Then they could implement new ideas in a fork, an one could run them side-by-side and migrate over to it.
Pluggable storage engines for databases don't work that well in practice. Either you end up with the MySQL situation where the storage engine is so dumb that you can't push any smart optimizations into it (making having pluggable engines moot in the first place), or you have to write such a large interface that it's not worth providing.
That depends on what you mean by pluggable storage and how it's implemented ...
For example PostgreSQL supported custom index access methods, which you might see as a custom storage format (although only for secondary storage). You had to modify the source code and rebuild PostgreSQL, but there was a fairly clear separation / internal API that allowed that. Since PostgreSQL 9.6 you can do that without the custom build (i.e. you can create a new index in an extension and use CREATE ACCESS METHOD to plug it into the server).
We don't have such clear internal separation for the primary storage, but it's hard to ignore the possible benefits of alternative storage types. Another reason is that we're constantly scavenging for free bits in various places (e.g. flags in tuple headers needed by new features etc), and allowing multiple formats would help with this by supporting "old" and "new" table format. So it'll likely follow what happened to indexes - build a clear internal API, allow multiple storage formats internally, eventually make it usable from extensions.
Yes. But it's not a settled matter. The storage engines in mysql didn't work out that well, each of them duplicating a lot of work, having significantly different behavior, ...
