JUSTICE BRENNAN delivered the opinion of the Court.
This case, commenced as a petition for an order to compel arbitration under § 4 of the United States Arbitration Act of 1925 (Arbitration Act or Act), 9 U. S. C. § 4, presents the question whether, in light of the policies of the Act and of our decisions in Colorado River Water Conservation District v. United States, 424 U.S. 800 (1976), and Will v. Calvert Fire Insurance Co., 437 U.S. 655 (1978), the District Court for the Middle District of North Carolina properly stayed this diversity action pending resolution of a concurrent state-court suit. The Court of Appeals for the Fourth Circuit reversed the stay. 656 F.2d 933, rehearing denied, 664 F.2d 936 (1981). We granted certiorari. 455 U.S. 937 (1982). We affirm.
Petitioner Moses H. Cone Memorial Hospital (Hospital) is located in Greensboro, N. C. Respondent Mercury Construction Corp. (Mercury), a construction contractor, has its principal place of business in Alabama. In July 1975, Mercury and the Hospital entered into a contract for the construction of additions to the Hospital building. The contract, drafted by representatives of the Hospital, included provisions for resolving disputes arising out of the contract or its breach. All disputes involving interpretation of the contract or performance of the construction work were to be referred in the first instance to J. N. Pease Associates (Architect), an independent architectural firm hired by the Hospital to design and oversee the construction project. With certain
The contract also specified the time limits for arbitration demands.
Construction on the project began in July 1975. Performance was to be completed by October 1979.
In January 1980, Mercury submitted to the Architect its claims for delay and impact costs. Mercury and the Architect discussed the claims over several months, substantially reducing the amount of the claims. According to the Hospital, it first learned of the existence of Mercury's claims in April 1980; its lawyers assumed active participation in the claim procedure in May. The parties differ in their characterizations of the events of the next few months — whether there were "ongoing negotiations," or merely an "investigation" by the Hospital. In any event, it appears from the record that lawyers for the Hospital requested additional information concerning Mercury's claims. As a result, on August 12, 1980, Mercury gave a detailed presentation of its claims at a meeting attended by Mercury's representatives and lawyers, the Hospital's representatives and lawyers, and representatives of the Architect. Mercury agreed to send copies of its files to an expert hired by the Hospital, and the parties agreed to meet again on October 13.
On October 6, Mercury's counsel telephoned the Hospital's counsel to confirm that the scheduled meeting would go forward. The Hospital's counsel said he would call back the next day. When he did, he informed Mercury's counsel that the Hospital would pay nothing on Mercury's claim. He also said that the Hospital intended to file a declaratory judgment action in North Carolina state court.
On October 15, without notice to Mercury, the Hospital obtained an ex parte injunction from the state court forbidding Mercury to take any steps directed toward arbitration. Mercury objected, and the stay was dissolved on October 27. As soon as the stay was lifted, Mercury filed the present action in the District Court, seeking an order compelling arbitration under § 4 of the Arbitration Act, 9 U. S. C. § 4.
Before we address the propriety of the District Judge's stay order, we must first decide whether that order was appealable to the Court of Appeals under 28 U. S. C. § 1291.
Mercury sought appellate review through two alternative routes — a notice of appeal under § 1291, and a petition for mandamus under the All Writs Act, 28 U. S. C. § 1651.
Idlewild Liquor Corp. v. Epstein, 370 U.S. 713 (1962), is instructive in this regard. There the plaintiff brought a federal suit challenging the constitutionality of a state statute. The District Judge declined to convene a three-judge court and stayed the federal suit under the Pullman abstention doctrine.
There can be no dispute that this order meets the second and third of these criteria. An order that amounts to a refusal to adjudicate the merits plainly presents an important issue separate from the merits.
The Hospital contends nevertheless that the District Court's stay order did not meet the first of the criteria, namely that it "conclusively determine the disputed question." But this is true only in the technical sense that every order short of a final decree is subject to reopening at the discretion of the district judge.
We turn now to the principal issue to be addressed, namely, the propriety of the District Court's decision to stay this federal suit out of deference to the parallel litigation brought in state court. Colorado River Water Conservation District v. United States, 424 U.S. 800 (1976), provides persuasive guidance in deciding this question.
Colorado River involved the effect of the McCarran Amendment, 66 Stat. 560, 43 U. S. C. § 666, on the existence and exercise of federal-court jurisdiction to adjudicate federal water rights, 28 U. S. C. § 1345. The Amendment waives the Government's sovereign immunity to permit the joinder of the United States in some state-court suits for the adjudication of water rights. In Colorado River, however, the Government proceeded in Federal District Court, bringing suit against some 1,000 nonfederal water users, seeking a declaration of the water rights of certain federal entities and Indian tribes. Shortly thereafter, a defendant in that suit
We began our analysis by examining the abstention doctrine in its various forms. We noted:
After canvassing the three categories of abstention, we concluded that none of them applied to the case at hand. 424 U. S., at 813-817.
Nevertheless, we held that the District Court's dismissal was proper on another ground — one resting not on considerations of state-federal comity or on avoidance of constitutional
We declined to prescribe a hard-and-fast rule for dismissals of this type, but instead described some of the factors relevant to the decision.
As this passage makes clear, the decision whether to dismiss a federal action because of parallel state-court litigation does not rest on a mechanical checklist, but on a careful balancing of the important factors as they apply in a given case, with the balance heavily weighted in favor of the exercise of jurisdiction. The weight to be given to any one factor may vary greatly from case to case, depending on the particular setting of the case. Colorado River itself illustrates this principle in operation. By far the most important factor in our decision to approve the dismissal there was the "clear federal policy . . . [of] avoidance of piecemeal adjudication of water rights in a river system," id., at 819, as evinced in the McCarran Amendment. We recognized that the Amendment represents Congress' judgment that the field of water rights is one peculiarly appropriate for comprehensive treatment in the forums having the greatest experience and expertise, assisted by state administrative officers acting under the state courts. Id., at 819-820. In addition, we noted that other factors in the case tended to support dismissal — the absence of any substantial progress in the federal-court litigation; the presence in the suit of extensive rights governed by state law; the geographical inconvenience of the federal forum; and the Government's previous willingness to litigate similar suits in state court. Id., at 820.
Before discussing the application of Colorado River's exceptional-circumstances test, we must address the Hospital's argument that that test was undermined by our subsequent decision in Will v. Calvert Fire Insurance Co., 437 U.S. 655 (1978). We find no merit in this argument for at least two reasons.
The Hospital further contends that Calvert requires reversal here because the opinions of JUSTICE REHNQUIST and
Applying the Colorado River factors to this case, it is clear that there was no showing of the requisite exceptional circumstances to justify the District Court's stay.
The Hospital concedes that the first two factors mentioned in Colorado River are not present here. There was no assumption by either court of jurisdiction over any res or property, nor is there any contention that the federal forum was any less convenient to the parties than the state forum. The remaining factors — avoidance of piecemeal litigation, and the order in which jurisdiction was obtained by the concurrent forums — far from supporting the stay, actually counsel against it.
There is no force here to the consideration that was paramount in Colorado River itself — the danger of piecemeal litigation.
The Hospital points out that it has two substantive disputes here — one with Mercury, concerning Mercury's claim for delay and impact costs, and the other with the Architect, concerning the Hospital's claim for indemnity for any liability it may have to Mercury. The latter dispute cannot be sent
The order in which the concurrent tribunals obtained and exercised jurisdiction cuts against, not for, the District Court's stay in this case. The Hospital argues that the stay was proper because the state-court suit was filed some 19 days before the federal suit. In the first place, this argument disregards the obvious reason for the Hospital's priority in filing. An indispensable element of Mercury's cause of action under § 4 for an arbitration order is the Hospital's refusal to arbitrate. See n. 27, infra. That refusal did not occur until less than a day before the Hospital filed its state suit. Hence, Mercury simply had no reasonable opportunity to file its § 4 petition first. Moreover, the Hospital succeeded in obtaining an ex parte injunction from the state court forbidding Mercury to take any steps to secure arbitration.
That aside, the Hospital's priority argument gives too mechanical a reading to the "priority" element of the Colorado River balance. This factor, as with the other Colorado River factors, is to be applied in a pragmatic, flexible manner with a view to the realities of the case at hand. Thus, priority should not be measured exclusively by which complaint was filed first, but rather in terms of how much progress has been made in the two actions. Colorado River illustrates
This refusal to proceed was plainly erroneous in view of Congress' clear intent, in the Arbitration Act, to move the parties to an arbitrable dispute out of court and into arbitration as quickly and easily as possible. The Act provides two parallel devices for enforcing an arbitration agreement: a stay of litigation in any case raising a dispute referable to arbitration, 9 U. S. C. § 3, and an affirmative order to engage in arbitration, § 4. Both of these sections call for an expeditious and summary hearing, with only restricted inquiry into factual issues.
Besides the four factors expressly discussed in Colorado River, there is another that emerges from Calvert — the fact that federal law provides the rule of decision on the merits. The state-versus-federal-law factor was of ambiguous relevance in Colorado River.
The basic issue presented in Mercury's federal suit was the arbitrability of the dispute between Mercury and the Hospital. Federal law in the terms of the Arbitration Act governs that issue in either state or federal court. Section 2 is the primary substantive provision of the Act, declaring that a written agreement to arbitrate "in any maritime transaction or a contract evidencing a transaction involving commerce. . . shall be valid, irrevocable, and enforceable, save upon such grounds as exist at law or in equity for the revocation of any contract." 9 U. S. C. § 2.
To be sure, the source-of-law factor has less significance here than in Calvert, since the federal courts' jurisdiction to enforce the Arbitration Act is concurrent with that of the state courts.
Finally, in this case an important reason against allowing a stay is the probable inadequacy of the state-court proceeding to protect Mercury's rights. We are not to be understood to impeach the competence or procedures of the North Carolina courts. Moreover, state courts, as much as federal courts, are obliged to grant stays of litigation under § 3 of the Arbitration Act.
The Hospital argues that the Colorado River test is somehow inapplicable because in this case the District Court merely stayed the federal litigation rather than dismissing the suit outright, as in Colorado River. It contends that Mercury remains free to seek to reopen the federal suit on a showing that the state suit has failed to adjudicate its rights, and that a stay is less onerous than a dismissal. We have already rejected this distinction, for purposes of this case, in discussing appellate jurisdiction. Supra, at 12-13. We reject it in this context for the same reasons.
Moreover, assuming that for some unexpected reason the state forum does turn out to be inadequate in some respect, the Hospital's argument fails to make out any genuine difference between a stay and a dismissal. It is true that Mercury could seek to return to federal court if it proved necessary; but that would be equally true if the District Court had dismissed the case. It is highly questionable whether this Court would have approved a dismissal of a federal suit in Colorado River (or in any of the abstention cases, see supra, at 14) if the federal courts did not remain open to a dismissed plaintiff who later demonstrated the inadequacy of the state forum.
In addition to reversing the District Court's stay, the Court of Appeals decided that the underlying contractual dispute between Mercury and the Hospital is arbitrable under the Arbitration Act and the terms of the parties' arbitration agreement. It reversed the District Court's judgment and remanded the case "with directions to proceed in conformity herewith." 656 F. 2d, at 946. In effect, the Court of Appeals directed the District Court to enter a § 4 order to arbitrate.
In this Court, the Hospital does not contest the substantive correctness of the Court of Appeals' holding on arbitrability. It does raise several objections to the procedures the Court of Appeals used in considering and deciding this case. In particular, it points out that the only issue formally appealed to the Court of Appeals was the propriety of the District Court's stay order. Ordinarily, we would not expect the Court of Appeals to pass on issues not decided in the District Court. In the present case, however, we are not disposed to disturb the court's discretion in its handling of the case in view of the special interests at stake and the apparent lack of any prejudice to the parties. Title 28 U. S. C. § 2106 gives a court of appeals some latitude in entering an order to achieve justice in the circumstances. The Arbitration Act calls for a summary and speedy disposition of motions or petitions to enforce arbitration clauses. The Court of Appeals had in the record full briefs and evidentiary submissions from both parties on the merits of arbitrability, and held that there were no disputed issues of fact requiring a jury trial before a § 4 order could issue. Under these circumstances, the court acted within its authority in deciding the legal issues presented in order to facilitate the prompt arbitration that Congress envisaged.
In its zeal to provide arbitration for a party it thinks deserving, the Court has made an exception to established rules of procedure. The Court's attempt to cast the District Court's decision as a final judgment fails to do justice to the meaning of the word "final," to the Act of Congress that limits the jurisdiction of the courts of appeals, or to the district judges who administer the laws in the first instance.
If the District Court had not stayed the proceeding, but had set a trial date two months away, there would be no doubt that its order was interlocutory, subject to review only by mandamus or pursuant to 28 U. S. C. § 1292(b). This would be true even though § 4 of the Arbitration Act provides that "the court shall proceed summarily" to trial, because an order setting a trial date only guides the course of litigation, and does not, of its own force, dispose of it on the merits. Such an order is tentative; that is, it is subject to change at any time on the motion of a party or by the court, sua sponte.
The order the District Court actually entered is no more final. It delayed further proceedings until the completion of pending litigation in the state courts. This order was also tentative; it was subject to change on a showing that the state proceedings were being delayed, either by the Hospital or by the court, or that the state courts were not applying the federal Act, or that some other reason for a change had arisen. This order did not dispose of the case on the merits. If the state court had found that there was no agreement to arbitrate within the meaning of the United States Arbitration Act, the District Court would have been bound by that finding. But res judicata or collateral estoppel would apply if the state court reached a decision before the District Court in the absence of a stay. The likelihood that a state court of competent jurisdiction may enter a judgment that may determine some issue in a case does not render final a federal district court's decision to take a two-day recess, or to order additional
Section 1291 of the Judicial Code is a congressional command to the federal courts of appeals not to interfere with the district courts' management of ongoing proceedings. Unless the high standards for a writ of mandamus can be met, or the district court certifies an interlocutory appeal pursuant to § 1292(b), Congress has directed that the district courts be permitted to conduct their cases as they see fit. The reason for this rule is simple:
The Court's decision places an unwarranted limitation upon the power of district courts to control their own cases. The Court's opinion does not establish a broad exception to § 1291, see ante, at 10-11, n. 11, but it does create uncertainty about when a district court order in a pending case can be appealed. This uncertainty gives litigants opportunities to disrupt or delay proceedings by taking colorable appeals from interlocutory
The Court has acknowledged the importance of the rule of finality as recently as Coopers & Lybrand v. Livesay, 437 U.S. 463 (1978), which rejected the so called "death knell" exception to § 1291. In Coopers, a putative representative plaintiff whose motion for class certification had been denied by the District Court sought to appeal under § 1291. We accepted his argument that this order effectively put him out of court, id., at 470, but held that this circumstance did not justify an exception to the statute. "[A]llowing appeals of right from nonfinal orders that turn on the facts of a particular case thrusts appellate courts indiscriminately into the trial process and thus defeats one vital purpose of the final-judgment rule — `that of maintaining the appropriate relationship between the respective courts. . . . This goal, in the absence of most compelling reasons to the contrary, is very much worth preserving.' " Id., at 476 (quoting Parkinson v. April Industries, Inc., 520 F.2d 650, 654 (CA2 1975) (concurring opinion)).
First, Mercury is less "effectively out of court" than was Idlewild. There was no pending state proceeding that might have resolved the issues in the case, and Idlewild might well have been obliged to take the risk of violating the statute and challenging it in an enforcement proceeding in state court.
More importantly, however, the decision in Idlewild cannot be good law after Coopers, supra. The Court describes Coopers as holding only that the collateral-order doctrine of Cohen v. Beneficial Loan Corp., 337 U.S. 541 (1949), does not apply to a class decertification order under Federal Rule of Civil Procedure 23(c)(1). Ante, at 12-13, n. 14. We did hold that "the collateral-order doctrine is not applicable to" a decertification order. 437 U. S., at 468-469. We then went on to reject the argument that the decertification order was final under the so-called "death knell" doctrine, holding that an order does not become final simply because the plaintiff will be unable to pursue his claim if the order stands. Id., at 469-477. We declined to attach any importance to the fact that the plaintiff in Coopers was just as "effectively out of court" as Idlewild or Mercury. We noted that "if the `death knell' doctrine has merit, it would apply equally to the many interlocutory orders in ordinary litigation . . . that may have such tactical economic significance that a defeat is tantamount to a `death knell' for the entire case." Id., at 470. We also noted that 28 U. S. C. § 1292(b) provides for review
The Court also states that the stay order in this case is appealable under Cohen, supra. It quotes the formulation of the Cohen collateral-order doctrine from Coopers:
The District Court's order did not "conclusively determine the disputed question" for the reasons stated above. The Court's assertion to the contrary, ante, at 12-13, is nothing short of sheer speculation about the state of mind of the District Judge. Such speculation is hardly the "practical rather than . . . technical construction"
For these reasons, I do not believe the District Court's order was appealable. Interlocutory orders are committed by statute to the judgment of the district courts, and this Court ill-serves the judges of those courts and the over-whelming majority of litigants by devising exceptions to the statute when it believes a particular litigant has been wronged.
Given my view of appealability, I do not find it necessary to decide whether the District Court's order was proper in this case. I am disturbed, however, that the Court has sanctioned an extraordinary departure from the usual and accepted course of judicial proceedings by affirming the Court of Appeals decision on an issue that was not decided in the District Court.
The Court of Appeals ordered the District Court to enter an order compelling arbitration, even though that issue was not considered by the District Court. This Court has maintained the difference between appellate jurisdiction and original jurisdiction at least since Marbury v. Madison, 1 Cranch 137, 174-176 (1803) ("It is the essential criterion of appellate jurisdiction, that it revises and corrects the proceedings in a case already instituted"). I do not understand how the Court can say that the Court of Appeals had discretion to perform a nonappellate act.
This statute does not grant the courts of appeals authority to constitute themselves as trial courts. Section 4 of the Arbitration Act gives the Hospital a right to a jury trial. See ante, at 23, n. 27. By deciding that there were no disputed issues of fact, the Court of Appeals seems to have decided a motion for summary judgment that was not before it. This is the kind of issue that district judges decide every day in the ordinary course of business. It is not the kind of issue that courts of appeals determine. The Court of Appeals did have before it the memoranda filed in the District Court but, contrary to the Court's intimation, ante, at 29, this issue was not argued in the Court of Appeals. See 656 F.2d 933, 948, n. 1 (1981) (Hall, J., dissenting) ("No one argued that this court should decide that issue").
There was no reason to believe that the District Court would not have acted promptly to resolve the dispute on the merits after being reversed on the stay. That judges of a court of appeals believe they know how a case should be decided is no reason for them to substitute their own judgment for that of a district judge without regard to the normal course of appellate procedure.
The judgment below should be vacated and the case remanded to the Court of Appeals with directions to dismiss the appeal for want of jurisdiction. Failing that, even if the Court is correct that the stay order was an error, the judgment should be reversed insofar as it decides the question of arbitrability, and the case should be remanded to the District Court for further proceedings under the Arbitration Act.
The contract also set a starting time limit for arbitration demands. No demand could be made earlier than 10 days after presentation of evidence to the Architect, unless the Architect rendered a written decision before that time. Id., at 29.
"The courts of appeals shall have jurisdiction of appeals from all final decisions of the district courts of the United States, . . . except where a direct review may be had in the Supreme Court."
There is no basis for the dissent's attempt, post, at 33, to distinguish Idlewild on the basis that in that case there was no pending state-court action when the District Court's stay issued. Neither the Court of Appeals nor this Court suggested in Idlewild that the state court's doors were anything but wide open to the plaintiff. "[E]ffectively out of court" means effectively out of federal court — in keeping with the fact that the decision under appeal is the refusal to exercise federal jurisdiction.
Moreover, the dissent's resolution of the appealability issue would yield the odd result that Pullman abstention orders would be immediately appealable in Texas but not in the other 49 States. Compare American Trial Lawyers Assn. v. New Jersey Supreme Court, 409 U.S. 467 (1973) (stays appropriate in Pullman cases), with Harris County Commissioners Court v. Moore, 420 U.S. 77, 88-89, and n. 14 (1975) (dismissal permissible to accommodate Texas jurisdictional requirements). This oddity illustrates the artificiality of resting appealability on an otherwise substanceless distinction between stays and dismissals in the present context. See Part IV-E, infra.
Of course, as these cases recognize, Idlewild does not disturb the usual rule that a stay is not ordinarily a final decision for purposes of § 1291, since most stays do not put the plaintiff "effectively out of court." See, e. g., Amdur, supra, at 105-106. Idlewild's reasoning is limited to cases where (under Colorado River, abstention, or a closely similar doctrine) the object of the stay is to require all or an essential part of the federal suit to be litigated in a state forum.
This answers the dissent's argument, post, at 33-34, that Idlewild was overruled by that part of Coopers & Lybrand v. Livesay, 437 U.S. 463, 469-477 (1978), which rejected the "death knell" doctrine of appealability. The "death knell" doctrine rested on the argument that in some situations an interlocutory decision (such as a refusal to certify a class) might terminate a suit as a practical matter because the named plaintiff would lack an economic incentive to pursue his individual claim. In a "death knell" case, however, the order sought to be appealed had no legal effect on the named plaintiff's ability to proceed with his individual claim in federal court. There is an obvious difference between a case in which the plaintiff himself may choose not to proceed, and a case in which the district court refuses to allow the plaintiff to litigate his claim in federal court. Appeal from a stay on abstention or Colorado River grounds, therefore, presents no prospect of "appeals of right from nonfinal orders that turn on the facts of a particular case," as in Coopers & Lybrand, supra, at 476. We foresee no great difficulty in determining when a district court has surrendered jurisdiction over a federal lawsuit.
For much the same reason, the dissent errs in likening the stay in this case to an ordinary delay in the interest of docket control, post, at 30-31. We do not hold that an order becomes final merely because it may have the practical effect of allowing a state court to be the first to rule on a common issue. We hold only that a stay order is final when the sole purpose and effect of the stay are precisely to surrender jurisdiction of a federal suit to a state court.
Coopers & Lybrand held that the Cohen rule did not apply to a class decertification order because, among other reasons, such an order is "inherently tentative" under Federal Rule of Civil Procedure 23(c)(1), which provides that such an order may be "altered or amended before the decision on the merits." 437 U. S., at 469, and n. 11. Of course, as Rule 54(b) provides, virtually all interlocutory orders may be altered or amended before final judgment if sufficient cause is shown; yet that does not make all pretrial orders "inherently tentative" in the sense of that phrase in Coopers & Lybrand. The rationale behind Rule 23(c)(1) is that a certification decision should be made "[a]s soon as practicable," even though later events or discoveries may mandate a different result. Many other orders, by contrast, are made with the expectation that they will be the final word on the subject addressed. Certainly that was true of the order at issue in this case. The reasoning of Coopers & Lybrand does not reach all pretrial orders that are formally subject to revision, but only those as to which some revision might reasonably be expected in the ordinary course of litigation.
The Court of Appeals in this case relied on similar reasoning. It concluded that, despite chronological priority of filing, the Hospital's state-court suit was a contrived, defensive reaction to Mercury's expected claim for relief and for arbitration. 656 F. 2d, at 944-945.
The reasoning of the Courts of Appeals in this case and in Calvert — that the vexatious or reactive nature of either the federal or the state litigation may influence the decision whether to defer to a parallel state litigation under Colorado River — has considerable merit. We need not rely on such reasoning here, however, for we conclude infra that even if the Hospital acted in complete good faith there were no exceptional circumstances warranting the District Court's stay.
In some cases, of course, it may be advisable to stay litigation among the nonarbitrating parties pending the outcome of the arbitration. That decision is one left to the district court (or to the state trial court under applicable state procedural rules) as a matter of its discretion to control its docket. See generally Landis v. North American Co., 299 U.S. 248, 254-255 (1936).
Section 4 provides that a district court must enter an order to arbitrate "upon being satisfied that the making of the agreement for arbitration or the failure to comply therewith is not in issue." If either of these points is in issue, § 4 provides that "the court shall proceed summarily" to a trial on that point. Section 6 further provides that a request for relief under either § 3 or § 4 is to be treated procedurally as a motion.
Moreover, the policy of the Arbitration Act requires a liberal reading of arbitration agreements, see infra, at 24-25. As a result, some issues that might be thought relevant to arbitrability are themselves arbitrable — further speeding the procedure under §§ 3 and 4. See, e. g., Prima Paint Corp. v. Flood & Conklin Mfg. Co., 388 U.S. 395 (1967).
The Arbitration Act is something of an anomaly in the field of federal-court jurisdiction. It creates a body of federal substantive law establishing and regulating the duty to honor an agreement to arbitrate, yet it does not create any independent federal-question jurisdiction under 28 U. S. C. § 1331 (1976 ed., Supp. V) or otherwise. Section 4 provides for an order compelling arbitration only when the federal district court would have jurisdiction over a suit on the underlying dispute; hence, there must be diversity of citizenship or some other independent basis for federal jurisdiction before the order can issue. E. g., Commercial Metals Co. v. Balfour, Guthrie, & Co., 577 F.2d 264, 268-269 (CA5 1978), and cases cited. Section 3 likewise limits the federal courts to the extent that a federal court cannot stay a suit pending before it unless there is such a suit in existence. Nevertheless, although enforcement of the Act is left in large part to the state courts, it nevertheless represents federal policy to be vindicated by the federal courts where otherwise appropriate.
We need not address whether a federal court might stay a state-court suit pending arbitration under 28 U. S. C. § 2283.