Split IDs into buckets to reduce Redis mem usage
This commit is contained in:
parent
b3b0654507
commit
cd4840363c
|
@ -1,4 +1,4 @@
|
|||
import asyncdispatch, times, strutils, tables
|
||||
import asyncdispatch, times, strutils, tables, hashes
|
||||
import redis, redpool, frosty, supersnappy
|
||||
|
||||
import types, api
|
||||
|
@ -22,7 +22,7 @@ proc migrate*(key, match: string) {.async.} =
|
|||
let list = await r.scan(newCursor(0), match, 100000)
|
||||
r.startPipelining()
|
||||
for item in list:
|
||||
if item != "p:":
|
||||
if item != "p:" or item == match:
|
||||
discard await r.del(item)
|
||||
await r.setk(key, "true")
|
||||
discard await r.flushPipeline()
|
||||
|
@ -34,14 +34,19 @@ proc initRedisPool*(cfg: Config) {.async.} =
|
|||
|
||||
await migrate("snappyRss", "rss:*")
|
||||
await migrate("oldFrosty", "*")
|
||||
await migrate("userBuckets", "p:")
|
||||
|
||||
pool.withAcquire(r):
|
||||
await r.configSet("hash-max-ziplist-entries", "1000")
|
||||
|
||||
except OSError:
|
||||
stdout.write "Failed to connect to Redis.\n"
|
||||
stdout.flushFile
|
||||
quit(1)
|
||||
|
||||
template toKey(p: Profile): string = "p:" & toLower(p.username)
|
||||
template toKey(l: List): string = toLower("l:" & l.username & '/' & l.name)
|
||||
template pidKey(name: string): string = "pid:" & $(hash(name) div 1_000_000)
|
||||
template profileKey(name: string): string = "p:" & name
|
||||
template listKey(l: List): string = toLower("l:" & l.username & '/' & l.name)
|
||||
|
||||
proc get(query: string): Future[string] {.async.} =
|
||||
pool.withAcquire(r):
|
||||
|
@ -52,23 +57,25 @@ proc setex(key: string; time: int; data: string) {.async.} =
|
|||
discard await r.setex(key, time, data)
|
||||
|
||||
proc cache*(data: List) {.async.} =
|
||||
await setex(data.toKey, listCacheTime, compress(freeze(data)))
|
||||
await setex(data.listKey, listCacheTime, compress(freeze(data)))
|
||||
|
||||
proc cache*(data: PhotoRail; name: string) {.async.} =
|
||||
await setex("pr:" & name, baseCacheTime, compress(freeze(data)))
|
||||
|
||||
proc cache*(data: Profile) {.async.} =
|
||||
if data.username.len == 0: return
|
||||
let name = toLower(data.username)
|
||||
pool.withAcquire(r):
|
||||
r.startPipelining()
|
||||
discard await r.setex(data.toKey, baseCacheTime, compress(freeze(data)))
|
||||
discard await r.hset("p:", toLower(data.username), data.id)
|
||||
discard await r.setex(name.profileKey, baseCacheTime, compress(freeze(data)))
|
||||
discard await r.hset(name.pidKey, name, data.id)
|
||||
discard await r.flushPipeline()
|
||||
|
||||
proc cacheProfileId*(username, id: string) {.async.} =
|
||||
if username.len == 0 or id.len == 0: return
|
||||
let name = toLower(username)
|
||||
pool.withAcquire(r):
|
||||
discard await r.hset("p:", toLower(username), id)
|
||||
discard await r.hset(name.pidKey, name, id)
|
||||
|
||||
proc cacheRss*(query: string; rss: Rss) {.async.} =
|
||||
let key = "rss:" & query
|
||||
|
@ -80,8 +87,9 @@ proc cacheRss*(query: string; rss: Rss) {.async.} =
|
|||
discard await r.flushPipeline()
|
||||
|
||||
proc getProfileId*(username: string): Future[string] {.async.} =
|
||||
let name = toLower(username)
|
||||
pool.withAcquire(r):
|
||||
result = await r.hget("p:", toLower(username))
|
||||
result = await r.hget(name.pidKey, name)
|
||||
if result == redisNil:
|
||||
result.setLen(0)
|
||||
|
||||
|
|
Loading…
Reference in New Issue