Oren Eini

CEO of RavenDB

a NoSQL Open Source Document Database

Get in touch with me:

oren@ravendb.net +972 52-548-6969

Posts: 7,503
Comments: 51,091
Privacy Policy · Terms
filter by tags archive
time to read 5 min | 890 words

Databases will typically divide their data files into pages, and manage all data in pages. As you can imagine, the notion of pages is pretty important to a database. Up to this point, we worked with pages, but not really. A page was just a buffer that we could read and write to. It took me 13 posts to manage to read and write pages from disk, but that is just the beginning. We need to add some structure for the pages. The usual way to do that is to have a page header, which contains some metadata about the page. That leads to an interesting question, how do we manage the page metadata?

One way to do that is to reserve a part of the page for this purpose, like so:


This is probably the simplest option (and what we use in Voron), but I don’t like it very much. It makes certain operations harder to deal with, because the usable size of the page is no longer a power of 2. Assuming we have 8KB pages and a page header of 64 bytes, that means that we can use just 8,128 bytes in the page. A lot of data structures are far easier to use if we can assume that the size is a power of 2.

Another thing to remember is that the underlying pager that we use operates on the level of 2 MB chunks. In other words, reading a value from the same 2MB range is as cheap as reading a value from the same 8KB page. For that reason, we can utilize the following approach, we’ll reserve the first section of the chunk for all the page headers. Here is what this will look like:


A chunk is 2MB in size, and pages are 8KB. That gives us 256 pages in a chunk. If we reserve the first page for page headers, that gives us 32 bytes for each page header. That is quite a lot, to be honest, which is great, because having more space on the header gives us a lot of options down the road.

An important consideration to pay attention to is that we read from the disk at 2MB chunks, but when we write, the writes happen at the page boundary. A change on Page #3, for example, means that we have to change both the header page and then page #3 as separate operations. If we were using an embedded page header, we’ll have just a single write to deal with. In practice, I don’t think this is a major issue. We are doing asynchronous writes and write coalescing anyway. For that matter, writes are actually done lazily (and the critical path goes to the Write Ahead Log anyway) so I don’t think this aspect matters so much.

At any rate, let’s get some idea about what the page header will look like:

I should note that this is merely a skeleton of the page header, just showing its structure. Over time, we’ll add a lot more details, but this is sufficient to understand the direction I’m going. A page header is denoted by a tag, which determines what type of page this is. Based on the type of the page, we can set various fields inside the header. You can also see the getPageCount(), which is a good demonstration of how we work with the page header. I need to have fine-grained control over the bytes in the header, so I’m setting them up in this manner.

Zig currently has a number of bugs related to packed / extern structs and unions, which make this a bit harder than expected, sadly. I’m waiting for the self hosted compiler, which will hopefully fix those issues.

With the PageHeader struct in place, we can start making use of this, like so:

There isn’t much going on here, which is great, but it does deserve some explanation. We first get the first page in the chunk for the page that was requested. For example, if we wanted to get page # 4, we get page #0, if we wanted to get page # 300, we’ll get page #256.  That page is the metadata page for the chunk, the only thing that it contains is the headers for all the pages in the chunk in question (including itself, of course).

The reason that this matters is that the header for the page is used to compute the number of pages that we need to return. As you’ll note, the API we have here does not provide a way for the caller to say how many pages should be returned. We need to figure it out on our own. To do that, however, we need to store that information somewhere. That is where the page header comes into place.

We get the page header of the page we care about from the metadata page of the chunk and compute the number of pages to get from the pager. Then we return to the caller the page header and the page buffer as a single unit.

Of course, this presupposes that the values got to the pages somehow. That leads us to other directions, however, dealing with page allocations. You can already see some hints of that in the code above, where we use freeSpaceBitmap. I’ll dedicate my next post to this topic.

time to read 13 min | 2570 words

In the last post, we figured out how we can find what pages we should flush to disk. In this one, I want to talk about how we can actually do the flushing. Once a page is marked as ready for flushing, we need to start an async process in which we write it to the disk. While that process is running, we need to ensure that there are no further writes happening on that page. We want to be able to handle flushing writes in the background with as little fuss as we possibly can, while at the same time maximizing parallelism. That turns out to be a significant challenge. Our previous write strategy simply gathered all the writes in memory and then flushed them to disk. Now we need to have an ongoing process. In addition, we also want to handle the scenario where we flush everything in memory for certain scenarios, such as checkpoints or orderly shutdown.

In short, the approach we previously took isn’t going to be sufficient for our needs. For that reason, I want to clear the board and start from scratch. As a reminder, here is the API that we provide:

The rules for this API are fairly simple. You acquire the writer and then may call the write() method any number of times. The writer will aggregate the writes in memory, and will use write behind to write the data to disk in the background. In other words, we aren’t necessarily going to write to the disk when write() is called, only to our own buffer.

Note that the writer is working using pages, not chunks, even though the buffer pool we use operates on chunks. In addition to the write() method, we have a flush() which pushes everything that is currently in the writer to the disk. That one is meant to be called when we want to create a checkpoint or when we do an orderly shutdown. The flush() call starts an asynchronous process that we can wait upon, so we aren’t going to stall on checkpoints.  While we write a page to disk, we’ll reject any writes to that page. This is important, because to do otherwise will mean that we may send messed up data to disk.

We also need to handle any and all sequences of writes. For example, consider:

  • write(2, [1 page]);  // write a single page #2
  • write(4, [4 pages]);// write 4 pages from page #4 (so pages: 4,5,6,7)

Now, we may actually get additional writes as well:

  • write(6, [3 pages]);
  • write(4, [1 page]);

In other words, there are no requirements from the caller about what should be called and in what order. The caller may write a set of pages and then repurpose them again. In this case, we expect that the end result is that we’ll issue the following writes to the disk:

  • write(2, 1 page);
  • write(4, 1 page);
  • write(6, 3 pages);

Note that we have page #5, it was written in the first call to page #4 (when it used 4 pages), however, with the update of page #4 to be a single page, the data in page #5 is no longer relevant and we can skip writing it to disk. This is an effect of working with pages. The data doesn’t make sense outside of a page boundary, after all, so we should only write meaningful data.

After spending a lot of time thinking about it, I decided to go with the following behavior: The Pager’s writer will manage the write buffers, flushing to the disk when it is time. Callers can decide to let writes happen in the background or to wait for them. Here is the new state of the Writer:

There are a few interesting fields here:

  • The lock is there to ensure that we have only a single writer. As a reminder, the actual I/O is asynchronous.
  • The cache implements an Lru with a twist, I improved on that since the last post, more on this later.
  • The parent field is the owning pager, obviously.
  • The last three fields (inflight, completed and backgroundWrites) are all about handling the asynchronous writes for us.

Let’s take a look at how we handle the state for a particular write and then we’ll be able to discuss the implementation in depth:

On the surface, the WriteState is pretty simple. It contains the page we are writing and how many pages are involved in a particular write, but there are also quite a few additional fields that are… strange. The writer field is used when we are returning from a callback, the chunks are used to record which chunks are relevant for this particular write (remember, the Writer operates at the page level, while the Pager operates at the chunk level). Because we deal with individual pages, we have to ensure that each one of them has its own reference for the chunk in question. That also has the effect of pinning any dirty pages in memory until they are flushed to disk. The err field holds whether there was an error (and what was it) when writing to the disk. The list field is used in conjunction with the write behind policy and will be discussed shortly. We use the next field to create a single linked list with the head being the completed field on the Writer.

With that in mind, let’s look at how the write() method is implemented:

We start by draining any completed writes. It is important to detect any errors and to release the references from the chunks we locked during the write portion. We treat any error during the write process as fatal and abort immediately. At this point, we don’t know what the state of the system is. The only safe way to handle this is to shutdown and run recovery.

The next step is ensureNotWrittenTo() call. This is required because we may be in the process of writing a page to the disk, and we cannot modify the memory for that page (or pages) while they are being written to the disk. This is a recoverable error, by the way. The caller should wait until the pending writes are complete and retry.

We then check in the cache if the page is already there. If it is, we have to consider the potential for a mismatch in the sizes (described earlier in this post), when we have a previous write that spanned three pages but the current one only applies to a single page. Since the page boundary is important, we can just discard the previous write and continue. For the common case of a write to a page that was previously written to, however, we can simply copy the data from the provided buffer to the right page in memory and move on. Most of the hot pages should fit into this code path.

We then allocate the state for the current write, get the page (or pages) and copy the data from the provided buffer to our own memory. We also, crucially, push the new state into the writer’s cache. In the previous post that action would evict a page that we need to handle. In the current code, that is handled at the epilog of the method.

We iterate over the cache and evict any pages that don’t fit into the cache any longer. This is because a single write may evict multiple pages. If we have space for 32KB, and we currently have 4 separate pages in the cache, we are full. Adding a write for 2 pages, however, will need to evict two items out of the cache. Another aspect that I changed from the last post is that I’m no longer requiring that the current page will not be evicted immediately.

I mentioned that I changed the Lru’s implementation significantly.  Previously it was fairly generic, but it turns out that it is better to specialize it specifically for our needs. Here is the current version:

This Lru is implemented using a generational approach. The idea is that we can specify the sizes of the young and old generations independently. A new page that is added to the cache is written to the young generation. If there is another write to the page, it will be moved to the old generation. If the size of the data in each generation is too big, we use a simple Least Recently Used to evict items. An item that expired on the old generation will be moved to the young generation, while an item that expires from the young generation will be evicted completely. Here is how this looks like in code:

This is the same rough algorithm used by MySQL. The advantage here is that there is O(1) cost for the entire process, but at the same time, busy pages actually have a good chance of not being written to the disk (which is ideal, since if they are busy, they keep changing). To give some context to the sizes involved, we may decide that the old generation contains 192 MB and the young generation is 64 MB. A page write will remain in the young generation until we have another 64 MB (8,192 individual page writes), if there were any writes to it again in that time frame, it will go to the old generation. Only if we didn't have any writes to it in that time frame will we flush it to disk. Once a page was moved to the old generation, it has to go through a buffer of 256 MB (32K page writes) before it will go to the disk. Hot pages should be safe from spurious writes, especially since we’ll write them to the disk anyway whenever we do a checkpoint.

When a page is evicted from the cache, the writePageToDisk() method is called on it, there is a bit of work going on here:

An important observation about writing to the disk is that coalescing writes can have a huge impact on performance. Instead of having to go to the disk multiple times, we can schlepp the data there just once. The amount of the data we write is actually far less significant than what you may expect. You can see that the first thing that we do is to try to merge the pages (both forward and backward). If we find a candidate, we expand our own WriteState and release the one that subsumed. This is an important aspect, because the writes we send to the disk don’t necessarily match the old / young generations division. If we evicted a page that is adjacent to another modified page, we’ll write them both to the disk. That is even if the second page is in the old generation (and that will evict the page from the write cache). In practice, databases exhibit a lot of locality, so I don’t expect that to be an actual issue.

The last thing we do before actually submitting the I/O operation is to register the write in the inflight record. This is basically a sorted list (by the first page for that write) which we’ll check on any future writes. We also increment the number of background writes we have so we’ll know to wait for them to complete. Note that if we fail to submit the I/O, we decrement the number of backgroundWrites, we use this value to wait for async operations (for example, during shutdown). And since such errors are probably fatal, we are heading that way soon.

The actual I/O happens in the background and we’ll call the callback function when that is done (either successfully or with an error). So far, all the code we saw was single threaded and protected by the lock mutex. The callback, on the other hand, is running without any such protections. Furthermore, the callback thread is running all the notifications we have. We want to do as little as possible there. Here is what I ended up with:

We register the completed write in the completed linked list and wake any pending waiters. Any actual processing, of course, is moved to the part of the process where we actually run under lock and don’t have to worry about concurrency and multi threading. The actual behavior around completing writes is implemented in the drainCompleted() call that is part of the write() call.

Even though we are under lock, the I/O completion may compete with us on the completed list, so we’re using lock free methods to iterate through the list of completed actions. The actual behavior is straightforward, we remove the write from the list of inflight writes and return an error if the write had a problem. Note that we have to deal with potentially nested errors as well.  Aside from calling this as part of the write call(), we may also want to explicitly wait for all the pending operations to complete. This is done using the waitForWrites() call:

We drain all the completed writes and then wait for any pending ones to complete as well. We repeat this until there are no more backgroundWrites. Note that we run this under the lock, so we know that there can never be additional writes happening.

In special circumstances, we may want to ask the writer to push everything to the disk, we do that by calling the checkpoint() method:

There is almost nothing there, I’m happy to say, we are almost done. We evict pages from the cache from the youngest to the oldest, using the same write coalescing as before. Given write locality, that is likely to produce good results. We complete the process by waiting for all those writes to complete, and we are done.

There is just one last topic that I want to cover: Shutting  down:

Shutting down turns out to be a fairly involved process. This is especially the case if we are shutting down because of an error. We need to wait until all the pending writes have been completed (to do otherwise is to invite Use-After-Free bugs). That is why we call the waitForWrites() until it completes successfully. At worst, each time this is called will process a single write.

On shutdown, we can’t really start writing to the disk (after all, we may be shutting down because of the disk). We just discard the data in this case. That is something that we are okay to do, because we assume that any important data can be restored when we run recovery. This is one of those cases where the fact that we are building a database makes our life easier. We don’t need to try very hard to persist to disk, we can assume that as a given.

I’m pretty happy with how all this turned out. The full code for the pager is now at around 1,500 lines of code. I think that at this point we are pretty much at or near what you’ll need to replace the mmap() based pager that I demonstrated earlier in this series. This is the sort of code that needs a metric ton of tests, of course, and a lot of actual experience throwing stuff at it. The nice thing about this approach is that this is entirely in our hands. We can now implement whatever policies we want. The actual behavior is more or less done, but we can play with the exact mechanics and what will trigger what at will. In some cases, that can give us a great deal of power and flexibility.

time to read 6 min | 1106 words

In the previous post I outlined some ideas about how to implement a more efficient write behind. The idea is that whenever we write pages to the Pager, we’ll not trigger an immediate write to the disk. Instead, we’ll keep the data in memory and only write to the disk when we hit a certain threshold. In many write scenarios, there are certain pages that are modified a lot (like the root page in a B+Tree) and pages that are modified rarely (a leaf page that got entries and will not be modified again). There is no point in writing the popular page to the disk, we’ll likely get another write to them shortly anyway. That calls to a Least Frequently Modified approach.

We don’t need to use a more complex approach, (like the Clock Sweep algorithm we use for the Pager), because we don’t have to deal with the same scenarios. There are not likely to be cases similar to scans, which throws a lot of complexities of buffer pool implementations. Writes operations are far more predictable in general and follow a pretty strict power law distribution. The task is simple: we have the list of pages that were modified, and at capacity, we’ll select some to send to the disk. The question is how to make that decision.

The simplest option is to go with the least recently used model. That is trivial to implement, The idea is that we have the following sequence of writes (assuming we have a capacity of 4 pages):

1, 2, 3, 2, 3, 1, 2, 3, 4, 2, 1, 2, 3, 4, 4, 2, 1, 6

In this case, the page that will be selected for eviction is #3, since it wasn’t modified the longest. The other alternative is to use least frequently used, in which case we have the following frequency table:

Page Usages
1 4
2 5
3 4
4 3
6 1

In this case, we’ll want to select page #4 for eviction. Since it is the one least used. (We don't consider #6 because it is the one we just inserted). I can make arguments for both sides, to be frank. It makes sense that the least frequently used is going to be the most relevant, right? The problem is that we need to also account for decaying usage over time. What do I mean by this?

We may have a page that is very hot, it gets used a lot for a certain period of time. After that point, however, it is no longer being written to, but because it was frequently used, it will take a long time to evict from the pool. A good example of such a scenario is when we have a B+Tree and we are inserting values in ascending orders. All the values for the tree are going to be placed in the same page, so if we have a lot of inserts, that page is going to be hot. Once it is full, however, we’ll start using another page as the target and then the page will reside in memory until some other page will have more usage.  A good discussion of least frequency used implementation is in this blog post.

A nice way to deal with the issue of decaying priorities over time in an LFU setup is to use the following formula to compute the priority of the pages:

The idea is that we compute the priority of the page based on the last access, so we are very close to the most recently used option. However, note that we compute the distance between accesses to the page. A page that is infrequently accessed will have low usages and a high delta sum. That will reduce its priority. Conversely, a page that is heavily used will have a low delta sum and high usage, so its value will be near the top.

Another option is to go with another clock sweep option. In this case, we use a simple least recently used model, but we keep count of the frequency of usages. In that case, if we have to evict a page that is heavily used, we can reduce its usages and give it another chance. The advantage here is that this is a far simpler model to work with, but gives roughly the same results. Another option we have is to use the midpoint insertion LRU.

There is also another consideration to take. The I/O cost isn’t linear. If I’m writing page #3 to disk, it is basically free from my perspective to write nearby pages. It is the same exact cost, after all, so why not do that?

We’ll need to write our own doubly linked list. The Zig’s standard library only contains a single linked list. It doesn’t take long to write such a data structure, but it is fun to do so. I absolutely get why implementing linked lists used to be such a common practice in interviews:

There isn’t really much to discuss here, to be honest. There is a bunch of code here, but it is fairly simple. I just had to implement a few operations. The code itself is straightforward. It is a lot more interesting when we see it being used to implement the LRU:

The push() method is where it all happens. We have two options here:

  • We have a page already inside the LRU. In that case, we increment its usage counter and move it to the front of the list.
  • This is a new page, so we have to add it to the LRU. If there is enough capacity, we can just add it to the front of the list and be done with it.

However, things get interesting when we are at capacity. At that point, we actually need to select a page to evict. How can we do that? We scan the end of the list (the oldest page) and check its usage. If it has more than a single usage, we half its usage counter and move it to the front. We continue to work on the tail in this manner. In essence, high usage counter will get reset rather quickly, but this will still give us a fairly balanced approach, with more popular pages remaining in the pool for longer.

When we evict a page, we can return it back to the caller, which can then write it to the disk. Of course, you probably don’t want to just write a single page. We need to check if we have additional pages nearby, so we can consolidate all of them at once to the disk.

I’ll touch on that in my next post.

time to read 5 min | 834 words

In the last blog post I presented the manner in which the Pager can write data to disk. Here is a reminder:

We acquire the writer (and a lock on it), call write on the pages we want to write on (passing the buffer to write on them), and finalize the process by calling flushWrites() to actually write the data to disk. As a reminder, we assume that the caller of the Pager is responsible for coordination. While we are writing to a specific page, it is the responsibility of the caller to ensure that there are no reads to that page.

The API above is intentionally simplistic , it doesn’t give us a lot of knobs to play with. But that is sufficient to do some fairly sophisticated things. One of the interesting observations is that we split the process of updating the data file into discrete steps. There is the part in which we are updating the in memory data, which allows other threads to immediately observe it (since they’ll read the new details from the Pager’s cache). Separately, there is the portion in which we write to the disk. The reason that I built the API in this manner is that it provides me with the flexibility to make decisions.

Here are some of the things that I can do with the current structure:

  • I can decide not to write the data to the disk. If the amount of modified pages is small (very common if I’m continuously modifying the same set of pages) I can skip the I/O costs entirely and do everything in memory.
  • Flushing the data to disk can be done in an asynchronous manner. In fact, it is already done in an asynchronous manner, but we are waiting for it to complete. That isn’t actually required.

The way the Pager works, we deposit the writes in the pager, and at some future point the Pager will persist them to disk. The durability aspect of a database is not reliant on the Pager, it is a property of the Write Ahead Log, usually.

If I wanted to implement a more sophisticated approach for writing to the disk, I could implement a least recently used cache for the written pages. When the number of pages in memory exceeds a certain size, we’ll start writing the oldest to disk. That keeps the most used pages in memory and avoids needless I/O.  At certain points, we can ask the Pager to flush everything to the disk, this gives us a checkpoint, where we can safely trim the Write Ahead Log. A good place to do that is whenever we reach the file size limit of the log and need to create a new one.

So far, by the way, you’ll notice that I’m not actually talking about durability, just writing to the disk. The durability aspect is coming from something we did long ago, but didn’t really pay attention to. Let’s look at how we are opening files, shall we:

Take a look at the flags that we pass to the open() command, we are asking the OS to use direct I/O (bypassing the buffer pool, since we’ll use our own) as well as using DSYNC write mode. The two together means that the write will skip any buffering / caching along the way and hit the disk in a durable manner. The fact that we are using async I/O means that we need to ensure that the buffers we write are not modified while we are saving them. As we currently have the API, there is a strong boundary for consistency. We acquire the writer, write whatever pages we need and flush immediately.

A more complex system would be needed to manage higher performance levels. The issue is that in order to do that, we have to give up a level of control. Instead of knowing exactly where something will happen, we can have a more sophisticated approach, but  we’ll need to be aware that we don’t really know at which point the data will be persisted.

At this point, however, there is a good reason to ask, do we even need to write durably? If we are limiting the consistency of the data to specific times requested by the caller (such as when we replace the Write Ahead Log), we can just call fsync() at the appropriate times, no? That would allow us to use buffered writes from most I/O.

I don’t think that this would be a good idea. Remember that we are using multiple files. If we’ll use buffered I/O and fsync(), we’ll need to issue multiple fsync() calls, which can be quite expensive. It also means higher memory usage on the system because of the file system cache, for data we determine is no longer in memory. It is simpler to use direct I/O for the whole thing, after all.

In the next post, I’m going to show how to implement a more sophisticated write-behind algorithm and discuss some of the implications of such a design.

time to read 5 min | 949 words

At long last, we are now at the point where we can write data back to the disk.  Before we can do that, however, we need to figure out what sort of writes we want to allow. The idea that I have in mind for the Pager is to follow the same path as Voron does. In other words, for writes, we can make the following assumptions:

  • There is only a single thread doing writes to the pager.
  • There are no readers for the writes until the write is completed.
  • There is a distinction between writing the data to the pager and writing the data to the disk.

Let’s break those assumptions apart and see what they bring to the table. The fact that we can assume only a single writer thread at any given point is pretty important. It means that the level of complexity that we have to face is greatly reduced. In the same sense, the fact that we don’t need to deal with concurrent readers or any consistency boundary for the data while it is being written will greatly simplify things for us. Finally, we make a distinction between writing to the pager and writing to the disk. Writing to the disk is _slow_, so we want to avoid doing that at any critical areas and push that to the background.

Finally, there is another aspect to consider. Internally, the Pager works with 2MB chunks, but to the outside world, it is using 8KB pages. When we write, we always write at the 8KB pages, not chunks. How would that work for the Pager?

The Pager itself is concurrent, but we only allow a single writer at a time, we can achieve this by centralizing all the write activities in the Writer struct, like so:

For now, I’m going to ignore the fields in the Writer struct, we’ll touch on them in detail later. In order to use the writer, you need to acquire it, write as many pages as you need, then release it. Here is a usage example:

The basic idea is fairly simple. With the writer, we operate at the page boundary to write as many pages as we need, once we are done, the call to flushWrites() persists the data to disk and then we can release the writer. Let’s dig a bit deeper and see how that works, shall we?

The write() call is about as basic as you can get. We use the getPage() function to get the right page, memcpy the data and that is about it, right? There are only two other things here that are important:

  • We record which chunk (the 2MB chunk of memory, out of which we carve the 8KB pages) at the writer’s level, is using the loadedChunksForWrites value.
  • We remember which pages we wrote to using the writtenPages hash table.

This is intentionally bare bones, because that is actually sufficient for our needs. The  fact that we remember which chunks we loaded (and keep a reference to them) will prevent us from reclaiming them, so even though we just wrote to memory, another thread can get the data and start using it without waiting for the disk. Of course, we still need to hit the disk eventually, that is what flushWrites() is about.

There is a lot that is going on here, let’s break it up. We start by allocating a temporary array and copying the keys from the writtenPages hash table to it. We then sort the array. This is done so we’ll be able to process the writes in a sequential manner, which is likely to be faster, even with async I/O. We then scan the list of pages in order, trying to merge writes together. The idea is to issue the minimum number of write calls. Finally, we’ll wait for all the writes to complete. Okay, maybe it isn’t that complex.  There is a bunch of code here, but it is mostly straightforward. Note that we also prepare the writeResults list to accept the results of the write to the disk.

As for writing to the disk, this is done using the PagerRing we previously looked at:

To write a buffer to the disk, we simply get the buffer from the Pager (reusing all the work we did in getPage()), increment the number of outstanding writes and then submit the work for the ring for processing. We setup the completeFlush as the callback function on completion. The PagerRing will call us when it is done writing to the disk. If there is an error, we’ll record it and reduce the number of outstanding writes. If there are no more outstanding writes, we’ll wake any waiters. That part is handled in the waitForAllDiskWritesToComplete().

We start by waiting for the outstanding writes to complete, waiting if needed. Then we can reset the state of the Writer. We start by resetting the written pages and then iterate over all the loaded chunks and release them. After the call, the Pager may decide to remove them from memory. This is fine, since they were already written to disk.

Except… if there was an error. You might have noticed that we are gathering the errors on each individual write operation we send, but we are actually only looking at the first one. For that matter, we clear the state of the Writer regardless if there were errors or not.

In general, an I/O error from the disk is not something that is recoverable. What you can do at this stage is to raise the error higher and run whatever recovery you have on startup.

In the next post, I’m going to be talking about durability and the overall expected performance of the system under this sort of model.

time to read 10 min | 1893 words

Up to this point, we focused on reading data from the disk, we can do that up to a point. Eventually we’ll run out of memory (assuming that the database is bigger than memory, which is a pretty safe assumption). That means that we need to decide what to remove from memory. When we use mmap(), the OS gets to decide that for us. In fact, that is probably the best argument against using mmap(). The additional control we get from knowing how to manage the memory is the chief reason to take the plunge and manage our own memory.

There are a lot of algorithms around managing memory, I really like this one, because it is quite elegant. However, that requires quite a lot of states to be dealt with, especially when working with highly concurrent systems. Instead, I chose to look at the clock sweep algorithm. This is also implemented by PostgreSQL, and it is actually far simpler to work with. The idea is that for each page, we maintain a usage count. Each time that we need to get a page, we’ll increment its usage count (up to a small limit). Each time we need to evict a page, we’ll search for a page that can be evicted and has no recent usage. If it has usages, we’ll decrement that value and repeat until we find something.

Our buffer management isn’t actually dealing with pages, however. We are working with 2MB chunks, instead. The principal is the same, but using bigger aggregates is advantageous given typical memory sizes these days. The first thing that we need to do is to modify the ChunkMetadata. I’m showing only the relevant changes.

The major change here is the introduction of the usages field. That is a 3 bits field (0 .. 8 in range) and we reduced the number of references a chunk can have to about 500 million (should be sufficient, I believe). The idea here is that each time that we call addRef(), we’ll increment the usages count, like so:

Zig has a nice feature that I’m using here, saturating addition. In other words, if the value is incremented beyond its limit, it is clamped to the limit. That means that I don’t have to worry about overflows, etc. I took a look at how this is implemented, and the compiler generates the following assembly for this code (x +| 100) :

This may look daunting, but I’ll break it down. First, we add the 100 to the value, as you can expect (the value is currently in the EAX register). Then we store –1 (value of 0xFFFFFFFF) in the ECX register. Finally, we use the CMOV instruction (the CMOVB in the snipper is a variant of CMOV), telling it to store ECX in EAX if the carry flag is set on the last addition instruction. For fun, this also avoids a branch in the code, which is great for performance.

One of the critical functions that we need to consider here is the behavior of the pager when we are accessing rare pages once, and then never again. A really common scenario is when we are doing a scan of some data for a rare query. For that reason, the usages behavior is a bit more complex than one might imagine. Let’s explore this for a bit before moving on. Look at the following code, I marked the important lines with stars:

When we mark a chunk as loaded, we copy the usages from the current record. That starts out as zero, so for the scenario of accessing rare pages, we’ll have a good reason to evict them soon. However, there is a twist, when we remove a chunk from memory, we also set its usage count to 1. That is an interesting issue. The chunk is not loaded, why does it have a usage count? That is because if we removed it from memory, and we load it again, we want it to start with a higher usage count (and less chance to be evicted). In this manner, we are somewhat simulating the 2Q algorithm.

Now, let’s take a look at the actual reclaiming portion, shall we? In the chunk metadata, we have the following behavior:

If we have a value and no outstanding references, we can reclaim it, if we don’t have a value, we’ll reduce the usage count anyway. The idea is that when we unload a page, we’ll set the usages count to 1. The usage counter of an unloaded page will reset this as time goes by, so if there has been a sweep on an empty chunk that has a usage count, we’ll not count that for the next time we load it. Basically, after unloading a chunk, if we reload it soonish, we’ll keep it around longer the next time. But if it is only rarely loaded, we don’t care and will forget that it was loaded previously.

The process for actually reclaiming a chunk is shown here:

We look at a particular index in the chunks, and check if we can claim that. Following out previous behavior, there are a bunch of options here. We can either have an empty chunk that remembers the previous usage, which we can reduce, or we can actually try to reclaim the chunk. Of course, that isn’t that simple, because even if we found a candidate for reclaiming, we still need to reduce its usages count. Only if it has no usages will we be able to actually start the removal process. Finally, we have the actual scanning of the chunks in the file, shown here:

We scan the file from the provided start index and try to reclaim each chunk in turn. That may reduce the usages count, as we previously discussed. The actual process will continue until we have found a chunk to reclaim. Note that we are always claiming just a single chunk and return. This is because the process is repeatable. We start from a given index and we return the last index that we scanned. If the caller needs to free more than a single chunk, they can call us again, passing the last index that we scanned.

That is why this is called the clock sweep algorithm. We are sweeping through the chunks that we have in the system, reaping them as needed. The code so far is all in the same FileChunks instance, but the Pager actually deals with multiple files. How would that work? We start by adding some configuration options to the pager, telling us how much memory we are allowed to use:

We have both soft and hard limits here, because we want to give the users the ability to say “don’t use too much memory, unless you really have to”.  The problem is that otherwise, users get nervous when they see 99% memory being used and want to keep some free. The point of soft and hard limits is that this gives us more flexibility, rather than setting a lower than needed limit and getting memory errors with GBs of RAM to spare.

In the Pager, we have the loadChunksToTransaction() that we looked at in the previous post. That is where we read the chunk from the file. We are going to modify this method so will reserve the memory budget before we actually allocate it, like so:

As you can see, we reserve the memory budget, then actually allocate the memory (inside markLoading()). If there is a failure, we release the budget allocation and report the error. To manage the memory budget, we need to add a few fields to the Pager:

You can see that releaseChunkMemoryBudget() is pretty trivial. Simply release the memory budget and move on. Things are a lot more complex when we need to reserve memory budget, however. Before we dive into this, I want to talk a bit about the filesRelcaimSweeps field. That is an interesting one. This is where we’ll keep the last position that we scanned in the Pager (across all pages). However, why is that an array?

The answer is simple. The Pager struct is meant to be used from multiple threads at the same time. Under memory pressure, we are likely to need to evict multiple chunks at once. In order to avoid multiple threads scanning the same range of the Pager to find chunks to remove, I decided that we’ll instead have several sweeps at the same time. On startup, we’ll initialize them to a random initial value, like so:

In this manner, under load, each thread is likely to scan an independent portion of the Pager’s memory, which should avoiding competing on the same memory to evict. And with that behind us, let’s see how we can actually use this to evict memory:

There is a lot of code here, and quite a few comments, because this is choke-full of behavior. Let’s dissect this in detail:

We start by checking the memory budget and see if we are below the soft memory limits. We are doing this optimistically, but if we fail, we reset the state and then start to scan the Pager for chunks we can release. We do that by accessing one of the filesRelcaimSweeps values using the current thread id. In this way, different threads are likely to use different values and move them independently.

We find the relevant file for the index and start scanning for chunks to release. We’ll stop the process when one of the following will happen:

  • We released a chunk, in which case we are successful and can return with glory.
  • We didn’t find a chunk, but found candidates (whose usage count is too high to discard).

In the case of the second option, we’ll look for better options before getting back to them and discarding them.

If we are completely unable to find anything to release, we’ll check if we exceed the hard memory limit and error, or just accept the soft limit as, well… soft limit and allocate the budget anyway.

This happens as part of the process for loading chunks in the pager, so we’ll only need to release a single chunk at a time. For that reason, we remember the state so the next operation will start from where we left off. You can think about this as a giant set of hands that are scanning the range of chunks in memory as needed.

There are actually a few things that we can implement that would make this faster. For example, we always scan through all the chunks in a file. We could try to maintain some data structure that will tell us which pages have usage count to consider, but that is actually complex (remember, we are concurrent). There is also another factor to consider. The ChunkMetdata is 64 bits in size, and a FilesChunk struct contains an array of 4096 such values, totaling 32KB in size. It is actually cheaper to scan through the entire array and do the relevant computation on each candidate than try to be smart about it.

I think that this is it for now, this post has certainly gone for quite a while. In the next post in the series, I want to tackle writes. So far we only looked at reads, but I think we have all the relevant infrastructure at hand already, so this should be simpler.

time to read 9 min | 1677 words

We have finally gotten to the point where we can ask the pager for a page of data (reminder, a page in this case is 8KB of data) and get it back. Here is what this looks like:

There are a few things to note here. There is a new concept here, the transaction. We don’t want to have that concept at the pager level, so we just pass the relevant state to the pager to deal with. Basically, any transaction we’ll have will need to have a bit of state for the pager to deal with it. The state we need at this point is minimal, just the list of chunks that are referenced by the pager. You can also see that we provide a timeout for the getPage() call. What is that about?

This leads to the second concern we need to consider. How are we expected to run this? If we call getPage() on a page (actually, a chunk containing this page) that isn’t resident in memory, we’ll need to go to the disk to read it. That can take a while, sometimes a long while. At a glance, that is one of those things that async/await was meant for. Since Zig supports async functions, that is certainly something that we can do, but it is something that I want to be cautious about. Having explicit blocking is far easier to understand and debug, at least for now. This is especially true if we’ll want to consume the pager API from a non Zig target.

That leads to an interesting issue, however. If a call to getPage() can block, how can we avoid blocking the thread. In most cases, we would like to avoid blocking, after all. It would be simple to have tryGetPage() method, which will not block (but schedule the load of the page from disk), and then maybe register for a notification for that. If that sounds like async to you, that is because it is. The problem with this sort of approach is that you need to suspend execution somewhere in the middle of a transaction operation and continue when the data is loaded. Without async/await, you can’t really do that. Well, I mean, you could try, but we have a lot of experience with trying to manage state via callbacks, that isn’t really going to work for anything beyond the simplest systems (see: node.js without async/await).

There is one thing that we can do that would be both simple and effective, however: we can error if the page isn’t in memory. That sounds like a pretty bad idea, no? How would that help us?

Well, the concept is simple. If a transaction attempts to access a page that isn’t resident in memory, we’ll do the following operations:

  • Schedule the chunk the page resides on to load into memory.
  • Return an error from the pager
  • Rollback the transaction
  • Keep the loadedChunks for that transaction active and wait for the chunk to be loaded
  • Re-run the transaction again, now the chunk is in memory and we can proceed further

Each time that we re-run the transaction, we make sure that the chunks it needs are in memory, eventually ensuring that all the required chunks are resident and we don’t need to block.

At the same time, the code to work with the transactions is not going to care about blocking, etc. We need to do the usual error handling, but that is required anyway. There is a single location where we need to deal with callbacks from the pager, so there is a limited blast radius of complexity. For write transactions, for example, this is a very reasonable strategy. We assume that there is only a single thread writing at a given time. A transaction being blocked because it needs to read a page from the disk can stall other pending transactions. By having it abort and retry later, we can keep the line moving. For read operations, on the other hand, that is likely not something that you want to do. If I’m already streaming results to the caller, I can’t just repeat the transaction.

I’m not making any decisions at this point, just considering the various options and implications that we have to deal with at this early level.

Now, let’s look at how the getPage() is actually implemented, shall we?

There is a lot that is going on here, I know. We start by defining a set (a hash map using a 64 bits unsigned integer to a zero-sized value). The way this works with Zig is quite elegant, since we pay no memory cost for the values here.

The majority of the work is done in the loadChunksToTransaction() function, which we’ll examine shortly, but you can see some interesting details already in getPage(). We assume that we have a page loaded, and any range of pages that we ask is always within a single page.

The call to load the chunks actually puts them in the loadedChuncks argument. We verify that we loaded the chunk properly and then we create a slice to return for the caller. Note that we may request more than a single page and it is valid to ask for a range that contains multiple chunks. We validate that the range we return is within the memory range for the current file, we ensured that the chunks for a specific file are consecutive in memory, so we can safely return this pointer across multiple chunks without needing to think about it.

There is another aspect of loadedChunks that we need to discuss. A transaction may use multiple pages from the same chunk, but we only need to load the chunk once. At the same time, we can avoid adding a reference to the chunk multiple times for each loaded page. When we close the transaction, we need to release the reference for these chunks, so we need to keep track of those. With that in mind, let’s see how we actually load the chunks to memory.

As a reminder, we have two actors working together here. The FileChunks is used to store the chunks in memory and the PagerRing is used for parallel I/O.

That is a lot of code to throw at you, I know, let’s dissect it in detail. In this method, we are working on chunks, not pages, and we assume that we may have multiple chunks, that is why we have the while loops. We start by checking if the chunk is already loaded in the transactions’ loadedChunks. If it isn’t, we compute the position of the chunk in the file (the chunk number we get from the caller is the global one, after all) and try to get it from the FileChunks. This is where things get interesting.  When we call tryGet() for the current chunk, we may get an error because of two possible scenarios:

  1. The value is currently being loaded from the disk (some other transaction asked for it, probably). We don’t need to do anything further other than wait for it to show up.
  2. Another transaction tried to load it, but got an error. At this point we just return the error. We don’t try to do anything special here. In general, there may be a lot of errors to consider here. We may have temporary I/O issue, or run out of memory or something that is transient. Or we may have an actual problem at hand (bad sector on disk, corrupted data, etc). Regardless of what we are doing, we aren’t going to try to do any error handling here. We’ll just record the error and any future attempt to access that chunk will also error. The proper way to recover at this point is to restart the pager. This is assuming we have the other components of a database at play here. So we’ll re-run the journal files, apply recovery, etc. In short, any I/O issues like that are critical errors and require a restart of the system to come back to a known state.

If the tryGet() method returned without an error, there are still two options to consider. The call may have returned a value (so we called addRef() on the chunk internally), we can simply add that to the chunks we own and move on. If there isn’t a value in memory, things start to get interesting. At this point we call markLoading(). We are basically racing to be the owners for loading this chunk. If we are successful in this race, we’ll get the buffer back from the FileChunks and can schedule reading the relevant chunk from the disk. You’ll note that we are setting the callback to completeLoad, we’ll look into that shortly. If we aren’t successful (we didn’t get a buffer back), then some other thread was able to get the buffer and will schedule the read for us, so we are done.

After we either ensured that all the chunks are loaded or scheduled them to be loaded, we use getBlocking() to wait for all the relevant chunks to be available. Once that is done, we can safely return and getPage() will complete the process, as we saw earlier.

The only thing that we have to look at is the completeLoad function, which is about as basic as you can get:

Most of the function is about error handling. We register either the fact that we got an error reading from the disk or that we completed the load process and maybe log something. In general, there isn’t really much that we need to do here. The act of calling markLoaded() will release any threads waiting on getBlocking(), after all. So the whole thing comes together quite nicely.

With this done, we are mostly done on the reading side of the pager and this post as well. In my next post, I want to discuss how we should handle eviction of data. So far, we are just reading into memory, never releasing. We need to take care of that as well, of course. Once that is done, we can move to the wonderful topic of handling writes and durability…

time to read 5 min | 970 words

This is my 7th post in this series, and we are now starting to get into the really interesting bits. So far we worked with the individual components, each of them doing just one thing, which makes it hard to see how they are all put together. In this post, we are going to start plugging things together.

As a reminder, the whole point of this series of posts is to explore what we need to do in order to ask for a page (8KB, in our case) from the data file and work with it in memory.  We have most of everything ready, let’s put them back together. The first thing that we need to do is actually tie a file to the FileChunks structure that we previously created. This is as simple as this structure:

We are simply initializing the FileChunks and opening the file, nothing else. Remember that a single pager is going to be responsible for multiple files. Furthermore, all the data structures that we are dealing with now are meant for concurrent use. That means that we should be prepared for this type of code:

When we use a managed language, that is actually fairly simple to work with. In an unmanaged language, those two lines of code are tough. Why is that? Let’s look at the raw data members for the Pager structure, shall we?

Of particular interest to us is the files member. This is the list of files that are being managed by the pager. Each one of them has a maximum size of 8GB in size. The problem is that we may have one thread accessing the list at the same time that another thread wants to increase its size. How would that work?

The simplest option is that we’ll reallocate the array, but that will move it. What would the first thread be doing in that scenario? The good thing from our point of view is that we don’t need to worry about concurrent modifications. There is only a single thread that is allowed to modify the Pager’s state at any given point in time.

Trying to find solutions for this problem leads into a rabid rabbit’s hole. You go into hazard pointers, epoch GC and other fun stuff. Also, the call to getPage() is one of the most important ones in a database, anything that we can do to reduce its cost will be done. As such, we can’t typically use any form of locking. A reader/writer lock can be a killer. Here is a good example for how that can happen.

I thought about how to resolve this, and I decided against a generic solution, instead, let’s look at the actual behavior that we need. The files array is going to be accessed a lot, it has an entry per 8GB of disk space that we take. That means that it isn’t going to be experiencing any rapid growth. It is also only going to grow. We also need to worry only when we grow the physical backing store for this array, if we overprovision and use less than we need, that is perfectly fine. Each element in the array is 8 bytes in size, so if we allocate a single memory page (4KB) we can store 512 file references in it. That represents 4 TB(!) of data, so we can probably just accept the additional cost and allocate it and not think about it.

Databases with > 4TB of disk size do exist, and we don’t want to have this artificial limit on us, do we? Instead, we can use another approach. We’ll start by allocating the array with a minimum size of 8 elements (sufficient until you get to 64GB). But what happens when we reach that size?

What we do here is cheat. We don’t need to free the memory immediately, when we reach the limit of the size of the array, we’ll double its size, copy the data that we have and register it to be freed when we close the Pager. At that point, the caller already needs to ensure that there are no other users of the pager.

Because we copy the values from the old array, but keep it around, old readers may use the old or new arrays, but we don’t actually care. The memory remains valid and accessible. In terms of wasted space, if our database went from a single file to being 128 TB in one run, we’ll have an array with 16,384 elements (whose size is 128KB). Along the way, we had to double the size of the array a dozen times and we “waste” 128KB of unused buffers. This seems like a pretty reasonable cost to significantly reduce the level of complexity of concurrent access. Using this method, we can avoid any sort of synchronization on the read side. That is certainly a plus.

Here are the init() and deinit() calls for the Pager, to complete the picture:

As you can see, we allocate a PagerRing for the pager, which will deal with the actual I/O. The actual disposal of the files array is managed using the pendingFree list. That is a small cost to pay, to reduce the cost of adding a new file. In the deinit() routine, note that there is a distinction between releasing the files themselves (where we close the FilesChunk, release the relevant memory, close the file handle, etc) and releasing the arrays that hold the files themselves.

I’m quite pleased with how this turned out, zero cost for reads (important) and negligible memory cost for most scenarios).

In my next post, I’ll get started with actually reading the data from disk and putting that in the pager. So far, that is a 7 post series, and we haven’t completed the first part. That simple scenario is surprisingly tricky.

time to read 5 min | 872 words

After implementing the memory management in the previous post, I set out to handle the actual I/O primitives that we need. As a reminder, we are separating the concerns here. We managed memory and reference counting in the previous post and now I want to focus on how we can read and write from the disk in as efficient a manner as possible. Before we get to the guts of the code, I want to explain a bit about the environment that I have in mind. Most of the time, the pager is able to provide the requested page from memory directly. If it can’t do that, it needs to consider the fact that there may be multiple threads that are trying to load that page.  At the same time, while we are loading the page, we want to be free to do other things as well.

I decided to implement the I/O routine using async I/O. Here is the rough sketch of the API I have in mind:

The idea is simple, we use a struct to which we can submit work in an asynchronous manner. At some later point in time, the work will complete and our read or write will be done. At that point we’ll be able to invoke the provided callback for the user. The code above is about as simple as you can manage, it spawns a dedicated thread to manage the I/O and then just issues those operations directly. To save myself some headache, I’m using an eventfd as a synchronization mechanism, I don’t strictly need this, but it will be useful down the road.

In terms of the API, I can now write the following code:

The basic idea is that the BackgroundRing struct doesn’t manage file descriptors or buffers. It is a strict way to manage I/O. The API is pretty piss poor as well, in terms of usability. No one will ever want to write generic I/O routines using this method, but we aren’t trying to do generic I/O, we are trying to ensure usability in a very constrained manner, inside the pager.

About the only nice thing that we do in this implementation is handle partial reads and writes. If we were asked to read more than what we got, we’ll repeat the operation until we get to the end of the file or succeed.

In terms of implementation, as well, this is a really bad idea. We look like we are doing async I/O, but we are actually just passing it all to a background thread that will do the work off a queue. That means that it will be very hard to make full use of the hardware capabilities. But I think that you can guess from the name that I’m not going to leave things like that. I’m going to use the new io_uring API in Linux to handle most of those concerns. That idea is that we’ll allocate a command buffer in the kernel and allow the kernel to handle asynchronous execution of the I/O operations. We still retrain the same rough structure, in that we are going to have a dedicated background thread to manage the commands, however. Amusing enough, the io_uring API is meant to be used from a single thread, since otherwise you’ll need to orchestrate writes to the ring buffer from multiple providers, which is much harder than a single consumer, single producer scenario.

The use of io_uring is also why we are using the eventfd model. We are registering that file descriptor in the io_uring so it will let us know when event completes. This also does double duty as the method that we can use to wake the background thread when we have more work for it to do. The most major change is inside the background worker, of course. Here is how this looks like:

We create  the ring in the init function (see full code listing below) and in the background thread we are simply waiting for an event using the eventfd. When a caller submits some work, we’ll register that on the io_uring and wait for it to complete (also using the eventfd). You can see that I’m handling some basic states (partial reads, full queues, etc). The code itself ends up being pretty small. Once we are done with the operation, we let the user know about the completion.

There are a few things here that are interesting to note. We are actually allowing interleaving of operations, so we may have many outstanding operations at any given point. We aren’t trying to guarantee any kind of ordering between the operations, nor are we providing anything but the most bare bones interface for the caller. Even so, there is quite a bit of power in the manner in which we are working here. We need to complete a few more components first, and then we can bring it all together…

Here is the full listing of the PagerRing code.  In the next post, I want to focus on the actual overall pager, when we are managing multiple files and how we work with all of them together. In particular, we want to understand how we manage the memory budget across all of them.

time to read 8 min | 1495 words

After writing the post about handling chunk metadata, I started thinking about the overall approach. Both the method using compressed pointers and the baseline computation felt… off to me. They were certainly workable, but it was too complex and felt fragile.

I don’t like dealing with a high level of complexity, I would rather put a lot of effort into simplifying the solution. The overall approach may be complex, but the system should be nice to work with. Usually, we can get away with a great deal of simplification if we accept some constraints on what we want to do with the system. For now, I’m going to assume the following constraints:

  • We are using 64 bits OS (and can assume effectively unlimited address space).
  • We want to go with a file pager (instead of the memory mapped one) because I want to be able to control the I/O behavior better.
  • The files we use are limited to 8 GB in size (can use more than a single file, of course).

The last one deserves some additional words. When thinking about a storage solution, accepting a maximum size is generally a bad idea (640KB, anyone?). However, if we decide that our storage solution is going to be composed of files of specific size, we can combine them to reach any size needed.

But why accept this limitation? Why say that a single file will not exceed 8 GB? It turns out that this has several advantages.

Let’s assume that we have a dataset that is 100GB in size, using 8 GB files, that would be 13 files to a total of 104 GB of used disk space. Now we want to delete some of that data. What do we do with the actual used disk space? It is actually quite hard to release disk space back to the operating system if you have a single file. You might need to run compaction of the data, or use advanced API such as hole punching (see FALLOC_FL_PUNCH_HOLE). Advanced API is something that I would like to avoid, too easy to fall into some pitfall that no one else has run into. Working with sparse files (with holes in them) also typically requires you to utilize dedicated tools and can be awkward.  If we split the data into separate files, we can retain most of the same benefits, and give ourselves a simpler environment for the user to work with.

With the 8GB limitation in place, I can choose to manage the paging using the following manner:

The idea is pretty simple. Instead of trying to stitch together the memory for the file, we are going to just allocate a single 8GB range of virtual memory. This can be done using the following command:

This reserves (but does not use) 8GB of address space. We can now allocate ranges from that safely. This is important because if we have a request to two sequential chunks, they will reside in memory right next to one another. Note that we also don’t need to handle any pointers, since we can rely on a stable base address for the whole file. The nice thing about this is that we aren’t actually allocating memory, just reserving it.

Let’s see how that will work? The chunks array is used to control references to the chunks in the file. The chunk metadata is a 64 bits value that has several responsibilities at the same time. It stores the tag of a chunk, which indicate its status (loaded, error, empty, etc) and the number of outstanding references to the chunk. That uses up 34 bits in the value, the rest of the bits are used as a version field, which is incremented on each change. That allows us to avoid the ABA problem. The actual data, of course, is managed using the ptr value.

Here is how we can get a chunk from this struct:

What we are doing here is checking that the value is loaded to memory, and if it is, we increment the reference and then return it. This code runs in a loop, because we assume that multiple threads may run it in the same time. This handles just getting data that is already loaded. If the data isn’t loaded, what will happen? We’ll get a null back. Here is the blocking version of this method:

Just based on those two methods, you should be able to draw some conclusions. If the value isn’t loaded, we’ll always return null, but there is this Loading stage as well, in that case, we may want to wait for it. How is that going to work?

This works using two important functions: markLoading() and markLoaded(), the idea is that we’ll first try to call tryGet() to load a chunk, if there is no value, we need to load it from disk. At that point, remember, there may be multiple threads accessing the relevant chunk. So all of them would be competing on the markLoading function, like so:

The code itself is pretty simple, we are updating the tag of the chunk and try to update it optimistically. We are moving the state of the chunk from Empty to Loading in a thread safe manner. If we are successful in doing so, we know that we are the only thread that owns the loading portion of the chunk. Note that part of the markLoading process is to ask the OS to give us the memory for the chunk (in the range that we previously allocated).

At this point, we can load the data from disk somehow and then we’ll call the markLoaded function, which completes the process:

The idea is that we are splitting the responsibility for managing the chunks references from how we load the data to memory.

In other words, the expected usage of this struct is something like this:

  1. Call tryGet() a page in a given chunk.
  2. If successful, do the work you wanted to do.
  3. If not successful, compete to be the loader for this data by calling markLoading().
  4. If you lost, call getBlocking() to wait for the winner to get the data.
  5. Somehow, load the data from the disk and call markLoaded().
  6. Proceed to make use of the data.

Another important aspect that we have to deal with is when we want to discard the data. Basically, if we filled our memory budget and we need to load a value from the disk, what can we do then? The answer is that we need to evict the data somehow, before we can do that, we need to know what data is currently in use. That is why we have the calls to addRef() and release(). We use those (using atomic operations) to track the usage of the various chunks. When we need to evict data from memory, we’ll need to have some sort of a policy to do so. I’m deferring the actual policy to a later point in time, right now I want to discuss how do we know what we can evict and how that is going to work.

Here is the code to handle eviction, currently implementing a policy of simple scanning (not ideal by a long shot):

In the reclaim method, we are scanning through the chunks. To be able to reclaim a chunk, the following conditions need to hold:

  1. The chunk holds a value.
  2. There are no outstanding references to the chunk, only the pager is holding a reference to the chunk.

Note that in order to do this safely, we have to assume that while we are trying to reclaim a chunk, another thread is trying to use it. This behavior complicates our lives a bit. We handle that by doing  a racy update of the chunk, trying to move it to a loading state. The idea is that the Loading state is meant to be used as a busy signal. While the chunk is in Loading state, the rest of the system knows that it cannot use this and needs to wait. Note that this means that we have the following transitions:


Most of the code that we have in the struct is there to handle concurrency from multiple threads dealing with the system at once, note. The actual behavior is fairly simple. We check if we can reclaim the chunk (no one is looking), we take a lock on by trying to move its state to Loading. Then we can discard the memory by calling mmap on the chunk’s memory with PROT_NONE.

For fun, we are using 2MB chunks because that fits well into huge pages. On a properly setup system, we can significantly reduce the paging metadata overhead inside the kernel by allocating a single 2MB page for each chunk.

You can see the entire implementation here. In the next post, I want to look into handling the I/O portion of reading the data from the disk. After that we’ll talk about how we can implement a proper eviction policy.


No future posts left, oh my!


  1. Recording (14):
    19 Jun 2024 - Building a Database Engine in C# & .NET
  2. re (33):
    28 May 2024 - Secure Drop protocol
  3. Meta Blog (2):
    23 Jan 2024 - I'm a JS Developer now
  4. Production postmortem (51):
    12 Dec 2023 - The Spawn of Denial of Service
  5. Challenge (74):
    13 Oct 2023 - Fastest node selection metastable error state–answer
View all series


Main feed Feed Stats
Comments feed   Comments Feed Stats