6
6
"database/sql"
7
7
"encoding/json"
8
8
"errors"
9
- "flag"
10
9
"fmt"
11
10
"io"
12
11
"net"
@@ -23,10 +22,9 @@ import (
23
22
24
23
"github.com/go-chi/chi/v5"
25
24
"github.com/google/uuid"
26
- "golang.org/x/exp/slices "
25
+ "golang.org/x/exp/maps "
27
26
"golang.org/x/mod/semver"
28
27
"golang.org/x/sync/errgroup"
29
- "golang.org/x/time/rate"
30
28
"golang.org/x/xerrors"
31
29
"nhooyr.io/websocket"
32
30
"tailscale.com/tailcfg"
@@ -39,7 +37,6 @@ import (
39
37
"github.com/coder/coder/v2/coderd/httpmw"
40
38
"github.com/coder/coder/v2/coderd/rbac"
41
39
"github.com/coder/coder/v2/coderd/util/ptr"
42
- "github.com/coder/coder/v2/coderd/util/slice"
43
40
"github.com/coder/coder/v2/codersdk"
44
41
"github.com/coder/coder/v2/codersdk/agentsdk"
45
42
"github.com/coder/coder/v2/tailnet"
@@ -1524,7 +1521,11 @@ func (api *API) workspaceAgentPostMetadata(rw http.ResponseWriter, r *http.Reque
1524
1521
key := chi .URLParam (r , "key" )
1525
1522
1526
1523
const (
1527
- maxValueLen = 32 << 10
1524
+ // maxValueLen is set to 2048 to stay under the 8000 byte Postgres
1525
+ // NOTIFY limit. Since both value and error can be set, the real
1526
+ // payload limit is 2 * 2048 * 4/3 <base64 expansion> = 5461 bytes + a few hundred bytes for JSON
1527
+ // syntax, key names, and metadata.
1528
+ maxValueLen = 2048
1528
1529
maxErrorLen = maxValueLen
1529
1530
)
1530
1531
@@ -1567,7 +1568,13 @@ func (api *API) workspaceAgentPostMetadata(rw http.ResponseWriter, r *http.Reque
1567
1568
slog .F ("value" , ellipse (datum .Value , 16 )),
1568
1569
)
1569
1570
1570
- err = api .Pubsub .Publish (watchWorkspaceAgentMetadataChannel (workspaceAgent .ID ), []byte (datum .Key ))
1571
+ datumJSON , err := json .Marshal (datum )
1572
+ if err != nil {
1573
+ httpapi .InternalServerError (rw , err )
1574
+ return
1575
+ }
1576
+
1577
+ err = api .Pubsub .Publish (watchWorkspaceAgentMetadataChannel (workspaceAgent .ID ), datumJSON )
1571
1578
if err != nil {
1572
1579
httpapi .InternalServerError (rw , err )
1573
1580
return
@@ -1593,7 +1600,42 @@ func (api *API) watchWorkspaceAgentMetadata(rw http.ResponseWriter, r *http.Requ
1593
1600
)
1594
1601
)
1595
1602
1596
- sendEvent , senderClosed , err := httpapi .ServerSentEventSender (rw , r )
1603
+ // We avoid channel-based synchronization here to avoid backpressure problems.
1604
+ var (
1605
+ metadataMapMu sync.Mutex
1606
+ metadataMap = make (map [string ]database.WorkspaceAgentMetadatum , 128 )
1607
+ // pendingChanges must only be mutated when metadataMapMu is held.
1608
+ pendingChanges atomic.Bool
1609
+ )
1610
+
1611
+ // Send metadata on updates, we must ensure subscription before sending
1612
+ // initial metadata to guarantee that events in-between are not missed.
1613
+ cancelSub , err := api .Pubsub .Subscribe (watchWorkspaceAgentMetadataChannel (workspaceAgent .ID ), func (_ context.Context , byt []byte ) {
1614
+ var update database.UpdateWorkspaceAgentMetadataParams
1615
+ err := json .Unmarshal (byt , & update )
1616
+ if err != nil {
1617
+ api .Logger .Error (ctx , "failed to unmarshal pubsub message" , slog .Error (err ))
1618
+ return
1619
+ }
1620
+
1621
+ log .Debug (ctx , "received metadata update" , "key" , update .Key )
1622
+
1623
+ metadataMapMu .Lock ()
1624
+ defer metadataMapMu .Unlock ()
1625
+ md := metadataMap [update .Key ]
1626
+ md .Value = update .Value
1627
+ md .Error = update .Error
1628
+ md .CollectedAt = update .CollectedAt
1629
+ metadataMap [update .Key ] = md
1630
+ pendingChanges .Store (true )
1631
+ })
1632
+ if err != nil {
1633
+ httpapi .InternalServerError (rw , err )
1634
+ return
1635
+ }
1636
+ defer cancelSub ()
1637
+
1638
+ sseSendEvent , sseSenderClosed , err := httpapi .ServerSentEventSender (rw , r )
1597
1639
if err != nil {
1598
1640
httpapi .Write (ctx , rw , http .StatusInternalServerError , codersdk.Response {
1599
1641
Message : "Internal error setting up server-sent events." ,
@@ -1603,97 +1645,57 @@ func (api *API) watchWorkspaceAgentMetadata(rw http.ResponseWriter, r *http.Requ
1603
1645
}
1604
1646
// Prevent handler from returning until the sender is closed.
1605
1647
defer func () {
1606
- <- senderClosed
1648
+ <- sseSenderClosed
1607
1649
}()
1608
1650
1609
- const refreshInterval = time .Second * 5
1610
- refreshTicker := time .NewTicker (refreshInterval )
1611
- defer refreshTicker .Stop ()
1651
+ // We send updates at most every second.
1652
+ const sendInterval = time .Second * 1
1653
+ sendTicker := time .NewTicker (sendInterval )
1654
+ defer sendTicker .Stop ()
1612
1655
1613
- var (
1614
- lastDBMetaMu sync.Mutex
1615
- lastDBMeta []database.WorkspaceAgentMetadatum
1616
- )
1617
-
1618
- sendMetadata := func (pull bool ) {
1619
- log .Debug (ctx , "sending metadata update" , "pull" , pull )
1620
- lastDBMetaMu .Lock ()
1621
- defer lastDBMetaMu .Unlock ()
1622
-
1623
- var err error
1624
- if pull {
1625
- // We always use the original Request context because it contains
1626
- // the RBAC actor.
1627
- lastDBMeta , err = api .Database .GetWorkspaceAgentMetadata (ctx , workspaceAgent .ID )
1628
- if err != nil {
1629
- _ = sendEvent (ctx , codersdk.ServerSentEvent {
1630
- Type : codersdk .ServerSentEventTypeError ,
1631
- Data : codersdk.Response {
1632
- Message : "Internal error getting metadata." ,
1633
- Detail : err .Error (),
1634
- },
1635
- })
1636
- return
1637
- }
1638
- slices .SortFunc (lastDBMeta , func (a , b database.WorkspaceAgentMetadatum ) int {
1639
- return slice .Ascending (a .Key , b .Key )
1640
- })
1641
-
1642
- // Avoid sending refresh if the client is about to get a
1643
- // fresh update.
1644
- refreshTicker .Reset (refreshInterval )
1645
- }
1646
-
1647
- _ = sendEvent (ctx , codersdk.ServerSentEvent {
1648
- Type : codersdk .ServerSentEventTypeData ,
1649
- Data : convertWorkspaceAgentMetadata (lastDBMeta ),
1650
- })
1651
- }
1652
-
1653
- // Note: we previously used a debounce here, but when the rate of metadata updates was too
1654
- // high the debounce would never fire.
1655
- //
1656
- // The rate-limit has its own caveat. If the agent sends a burst of metadata
1657
- // but then goes quiet, we will never pull the new metadata and the frontend
1658
- // will go stale until refresh. This would only happen if the agent was
1659
- // under extreme load. Under normal operations, the interval between metadata
1660
- // updates is constant so there is no burst phenomenon.
1661
- pubsubRatelimit := rate .NewLimiter (rate .Every (time .Second ), 2 )
1662
- if flag .Lookup ("test.v" ) != nil {
1663
- // We essentially disable the rate-limit in tests for determinism.
1664
- pubsubRatelimit = rate .NewLimiter (rate .Every (time .Second * 100 ), 100 )
1665
- }
1666
-
1667
- // Send metadata on updates, we must ensure subscription before sending
1668
- // initial metadata to guarantee that events in-between are not missed.
1669
- cancelSub , err := api .Pubsub .Subscribe (watchWorkspaceAgentMetadataChannel (workspaceAgent .ID ), func (_ context.Context , _ []byte ) {
1670
- allow := pubsubRatelimit .Allow ()
1671
- log .Debug (ctx , "received metadata update" , "allow" , allow )
1672
- if allow {
1673
- sendMetadata (true )
1674
- }
1675
- })
1656
+ // We always use the original Request context because it contains
1657
+ // the RBAC actor.
1658
+ md , err := api .Database .GetWorkspaceAgentMetadata (ctx , workspaceAgent .ID )
1676
1659
if err != nil {
1660
+ // If we can't successfully pull the initial metadata, pubsub
1661
+ // updates will be no-op so we may as well terminate the
1662
+ // connection early.
1677
1663
httpapi .InternalServerError (rw , err )
1678
1664
return
1679
1665
}
1680
- defer cancelSub ()
1666
+ for _ , datum := range md {
1667
+ metadataMap [datum .Key ] = datum
1668
+ }
1681
1669
1682
1670
// Send initial metadata.
1683
- sendMetadata (true )
1671
+
1672
+ var lastSend time.Time
1673
+ sendMetadata := func () {
1674
+ lastSend = time .Now ()
1675
+ metadataMapMu .Lock ()
1676
+ values := maps .Values (metadataMap )
1677
+ pendingChanges .Store (false )
1678
+ metadataMapMu .Unlock ()
1679
+ _ = sseSendEvent (ctx , codersdk.ServerSentEvent {
1680
+ Type : codersdk .ServerSentEventTypeData ,
1681
+ Data : convertWorkspaceAgentMetadata (values ),
1682
+ })
1683
+ }
1684
+
1685
+ sendMetadata ()
1684
1686
1685
1687
for {
1686
1688
select {
1687
- case <- senderClosed :
1689
+ case <- sendTicker .C :
1690
+ // We send an update even if there's no change every 5 seconds
1691
+ // to ensure that the frontend always has an accurate "Result.Age".
1692
+ if ! pendingChanges .Load () && time .Since (lastSend ) < time .Second * 5 {
1693
+ continue
1694
+ }
1695
+ sendMetadata ()
1696
+ case <- sseSenderClosed :
1688
1697
return
1689
- case <- refreshTicker .C :
1690
1698
}
1691
-
1692
- // Avoid spamming the DB with reads we know there are no updates. We want
1693
- // to continue sending updates to the frontend so that "Result.Age"
1694
- // is always accurate. This way, the frontend doesn't need to
1695
- // sync its own clock with the backend.
1696
- sendMetadata (false )
1697
1699
}
1698
1700
}
1699
1701
@@ -1717,6 +1719,10 @@ func convertWorkspaceAgentMetadata(db []database.WorkspaceAgentMetadatum) []code
1717
1719
},
1718
1720
})
1719
1721
}
1722
+ // Sorting prevents the metadata from jumping around in the frontend.
1723
+ sort .Slice (result , func (i , j int ) bool {
1724
+ return result [i ].Description .Key < result [j ].Description .Key
1725
+ })
1720
1726
return result
1721
1727
}
1722
1728
0 commit comments