Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
131 changes: 131 additions & 0 deletions di/cache/cache.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,131 @@
# Cache

`cache.q` provides an in-memory, parameterized caching mechanism for storing and reusing function results, reducing computation time for repeat calls.

## Configuration Variables

- **`.cache.maxsize`** — Maximum total cache size in MB.
- **`.cache.maxindividual`** — Maximum size in MB allowed for a single cache entry; capped at `maxsize`.
- **`MB`** — Defines one megabyte as `2 * xexp 20`.

## Core Structures

- **`cache`** (table) — Tracks cache entries:
- `id` (long)
- `lastrun`, `lastaccess` (timestamps)
- `size` (bytes)
- **`funcs`** (dict) — Maps `id` to the cached function.
- **`results`** (dict) — Maps `id` to the resulting data.
- **`perf`** (table) — Logs cache performance with columns:
- `time` (timestamp)
- `id` (long)
- `status` (symbol: `add`, `hit`, `fail`, `evict`, `rerun`)

## Main Functions

### `getid`
Generates unique IDs for new cache entries by incrementing a global counter.

### `add`
Takes parameters `[function; id; status]` and:
1. Executes `function` via `value`.
2. If result size ≤ `.cache.maxindividual * MB`, ensures enough space:
- Calculates required space and evicts older entries as needed.
3. Inserts or updates cache table, `funcs`, `results`, logs performance.
4. Otherwise, logs a `fail` and returns the result without caching.

### `drop`
Removes specific cache entries by `id`, updating both the cache table and results dict.

### `evict`
Evicts least-recently-accessed items until required space is freed:
- Sorts by `lastaccess`, sums sizes, iteratively drops entries.
- Logs `evict` in `perf`.

### `trackperf`
Logs performance events (`add`, `hit`, `fail`, `evict`, `rerun`) with timestamps into `perf`.

### `execute`
Parameters: `[func; age]`.
1. Looks for matching cache entry by function identity.
2. If found and `age <= now – lastrun`:
- Updates `lastaccess`, logs a `hit`, returns cached result.
3. If found but stale:
- Drops entry, logs `rerun`, re-executes via `add`.
4. If not present:
- Adds a new cache entry via `add`.

### `getperf`
Returns `perf` table with function mappings added for each event entry.

# Cache Example Usage

This example demonstrates how `.cache.execute` works with caching and stale time logic, along with performance tracking using `.cache.getperf[]`.

## Example Steps

### 1. First Execution
The function is run and the result placed in the cache:

```q
q) \t r:execute[({system"sleep 2"; x+y};1;2);0D00:01]
2023
q)r
3
```

### 2. Second Execution (Cache Hit)
The second time round, the result set is returned immediately from the cache as we are within the stale time value:

```q
q) \t r1:execute[({system"sleep 2"; x+y};1;2);0D00:01]
0
q)r1
3
```

### 3. Execution After Stale Time (Re-run)
If the time since the last execution is greater than the required stale time, the function is re-run, the cached result is updated, and the result returned:

```q
q) \t r2:execute[({system"sleep 2"; x+y};1;2);0D00:00]
2008
q)r2
3
```

### 4. Cache Performance Tracking
The cache performance is tracked using `.cache.getperf[]`:

```q
q).cache.getperf[]
time id status function
------------------------------------------------------------------
2013.11.06D12:41:53.103508000 2 add {system"sleep 2"; x+y} 1 2
2013.11.06D12:42:01.647731000 2 hit {system"sleep 2"; x+y} 1 2
2013.11.06D12:42:53.930404000 2 rerun {system"sleep 2"; x+y} 1 2
```


---

## Cache Table Schema

| Column | Type | Description |
|--------------|------------|--------------------------------------------------|
| `id` | `long` | Unique identifier for cached entry |
| `lastrun` | `timestamp`| When the entry was initially added |
| `lastaccess` | `timestamp`| When entry was last served from cache |
| `size` | `long` | Byte size of the cached result |

## perf Table Schema

| Column | Type | Description |
|---------|-------------|--------------------------------------------|
| `time` | `timestamp` | When the cache event occurred |
| `id` | `long` | Corresponding cache entry ID |
| `status`| `symbol` | Event type (`add`, `hit`, `fail`, etc.) |
| `function` (added via `getperf`) | `function` | Cached function for the event |

---

94 changes: 94 additions & 0 deletions di/cache/cache.q
Original file line number Diff line number Diff line change
@@ -0,0 +1,94 @@
/ Library to provide a mechanism for storing function results in a cache and returning them from the cache if they are available and non stale.

/ return timestamp function
cp:{.z.p};

/ the maximum size of the cache in MB
maxsize:10;

/ the maximum size of any individual result set in MB
maxindividual:50;

/ make sure the maxindividual isn't bigger than maxsize
maxindividual:maxsize&maxindividual;

MB:2 xexp 20;

/ a table to store the cache values in memory
cache:([id:`u#`long$()] lastrun:`timestamp$();lastaccess:`timestamp$();size:`long$());

/ a dictionary of the functions
.z.M.funcs set (`u#`long$())!();
/ the results of the functions
results:(`u#`long$())!();

/ table to track the cache performance
perf:([]time:`timestamp$();id:`long$();status:`symbol$());

id:0j;
getid:{:id+::1};

/ add to cache
add:{[function;id;status]
/ Don't trap the error here - if it throws an error, we want it to be propagated out
res:value function;
$[(maxindividual*MB)>size:-22!res;
/ check if we need more space to store this item
[now:cp[];
if[0>requiredsize:(maxsize*MB) - size+sum exec size from cache; evict[neg requiredsize;now]];
/ Insert to the cache table
.z.M.cache upsert (id;now;now;size);
/ and insert to the function and results dictionary
funcs[id]:enlist function;
results[id]:enlist res;
/ Update the performance
trackperf[id;status;now]];
/ Otherwise just log it as an addfail - the result set is too big
trackperf[id;`fail;cp[]]];
/ Return the result
res};

// Drop some ids from the cache
drop:{[ids]
ids,:();
delete from .z.M.cache where id in ids;
results:: ids _ results;
}

// evict some items from the cache - need to clear enough space for the new item
// evict the least recently accessed items which make up the total size
// feel free to write a more intelligent cache eviction policy !
evict:{[reqsize;currenttime]
r:select from
(update totalsize:sums size from `lastaccess xasc select lastaccess,id,size from cache)
where prev[totalsize]<reqsize;
drop[r`id];
trackperf[r`id;.z.M.evict;currenttime];
}



trackperf:{[id;status;currenttime] .z.M.perf insert ((count id)#currenttime;id;(count id)#status)};


// check the cache to see if a function exists with a young enough result set
execute:{[func;age]
// check for a value in the cache which we can use
$[count r:select id,lastrun from cache where .z.M.funcs[id]~\:enlist func;
// There is a value in the cache.
[r:first r;
// We need to check the age - if the specified age is greater than the actual age, return it
// else delete it
$[age > (now:.z.p) - r`lastrun;
// update the cache stats, return the cached result
[update lastaccess:now from .z.M.cache where id=r`id;
trackperf[r`id;`hit;now];
first results[r`id]];
// value found, but too old - re-run it under the same id
[drop[r`id];
add[func;r`id;`rerun]]]];
// it's not in the cache, so add it
add[func;getid[];`add]]}

// get the cache performance
getperf:{update function:funcs[id] from perf}
9 changes: 9 additions & 0 deletions di/cache/init.q
Original file line number Diff line number Diff line change
@@ -0,0 +1,9 @@
/ Load core functionality into root module namespace
\l ::cache.q

export:([
getperf:getperf;
sdd:add;
drop:drop;
execute:execute
])
40 changes: 40 additions & 0 deletions di/cache/test.csv
Original file line number Diff line number Diff line change
@@ -0,0 +1,40 @@
action,ms,bytes,lang,code,repeat,minver,comment
before,0,0,q,cac:use`di.cache,1,1,load package into session

/ Test 1: Baseline Addition with Sleep
run,0,0,q,system "t cac.execute[({system\"sleep 1\"; x+y};10;20);0D00:01]",1,1,Initial run with 1s sleep
true,0,0,q,0~ system "t cac.execute[({system\"sleep 1\"; x+y};10;20);0D00:01]",1,1,Speedup: Cached result returns in <1ms

/ Test 2: String Manipulation Speed
run,0,0,q,system "t cac.execute[({system\"sleep 1\"; upper x};\"hello world\");0D00:01]",1,1,Initial run with string conversion
true,0,0,q,0~ system "t cac.execute[({system\"sleep 1\"; upper x};\"hello world\");0D00:01]",1,1,Speedup: String conversion cached

/ Test 3: Matrix Multiplication (Mock)
run,0,0,q,system "t cac.execute[({system\"sleep 1\"; (reverse x) * y};500;1000);0D00:01]",1,1,Initial run with numeric calc
true,0,0,q,0~ system "t cac.execute[({system\"sleep 1\"; (reverse x) * y};500;1000);0D00:01]",1,1,Speedup: Numeric calc cached

/ Test 4: Join Operations
run,0,0,q,system "t cac.execute[({system\"sleep 1\"; x uj y};([]a:1 2);([]a:3 4));0D00:01]",1,1,Initial run with table join
true,0,0,q,0~ system "t cac.execute[({system\"sleep 1\"; x uj y};([]a:1 2);([]a:3 4));0D00:01]",1,1,Speedup: Table join cached

/ Test 5: Type Checking & Casting
run,0,0,q,system "t cac.execute[({system\"sleep 1\"; `int$x};123.456);0D00:01]",1,1,Initial run with casting
true,0,0,q,0~ system "t cac.execute[({system\"sleep 1\"; `int$x};123.456);0D00:01]",1,1,Speedup: Casting cached

/ Test 6: Nested List Flattening
run,0,0,q,system "t cac.execute[({system\"sleep 1\"; raze x};(1 2;3 4;5 6));0D00:01]",1,1,Initial run with raze
true,0,0,q,0~ system "t cac.execute[({system\"sleep 1\"; raze x};(1 2;3 4;5 6));0D00:01]",1,1,Speedup: Raze cached

/Test 7: Check if output is correct
run,0,0,q,cac.execute[({x+1};1);0D00:01],1,1,cache sleep functionality
true,0,0,q,2~cac.execute[({x+1};1);0D00:01],1,1,cached sleep functionality

/ Test 8: Table Result Correctness
run,0,0,q,res::([]a:1 2 3;b:`x`y`z);cac.execute[({x};res);0D00:01],1,1,Initial run: cache a table
true,0,0,q,res ~ cac.execute[({x};res);0D00:01],1,1,Check: Table retrieved from cache matches original

/ Test 9: Complex Dictionary Result
run,0,0,q,dict::`stats`data!((avg;med);til 10);cac.execute[({x};dict);0D00:01],1,1,Initial run: cache a dictionary of functions and lists
true,0,0,q,dict ~ cac.execute[({x};dict);0D00:01],1,1,Check: Dictionary retrieved from cache matches original